chore: integrate develop into hare/develop candidate

This commit is contained in:
2026-09-05 04:15:47 +09:00
56 changed files with 9246 additions and 2921 deletions
+6 -3
View File
@@ -40,7 +40,7 @@ use tracing_subscriber::EnvFilter;
use agen::{
Engine, EngineRunExit, RunInterruptionReason,
interceptor::{Interceptor, PostToolAction, ToolResultInfo},
interceptor::{Interceptor, InterceptorResult, PostToolAction, ToolResultInfo},
llm_client::{
LlmClient,
capability::{CacheStrategy, ModelCapability, StructuredOutput, ToolCallingSupport},
@@ -280,7 +280,10 @@ impl ToolResultPrinterPolicy {
#[async_trait]
impl Interceptor for ToolResultPrinterPolicy {
async fn post_tool_call(&self, info: &mut ToolResultInfo) -> PostToolAction {
async fn post_tool_call(
&self,
info: &ToolResultInfo<'_, ()>,
) -> InterceptorResult<PostToolAction> {
let name = self
.call_names
.lock()
@@ -294,7 +297,7 @@ impl Interceptor for ToolResultPrinterPolicy {
println!(" Result ({}): ✅ {}", name, info.result.summary);
}
PostToolAction::Continue
Ok(PostToolAction::Continue)
}
}
+336 -106
View File
@@ -15,8 +15,12 @@ use crate::{
},
handler::{ErrorKind, StatusKind, ToolUseBlockStart, UsageKind},
interceptor::{
DefaultInterceptor, Interceptor, PostToolAction, PreRequestAction, PreToolAction,
PromptAction, ToolCallInfo, ToolResultInfo, TurnEndAction,
AssistantTurnEndContext, DefaultInterceptor, Interceptor, InterceptorCallId,
InterceptorCounter, InterceptorCounters, InterceptorError, InterceptorErrorCategory,
InterceptorFailure, InterceptorInvocation, InterceptorPhase, InterceptorRunId,
InterceptorTurnId, PendingHistoryAppendsContext, PostToolAction, PreLlmRequestContext,
PreRequestAction, PreToolAction, PromptAction, PromptSubmitContext, RunExitContext,
ToolCallInfo, ToolResultInfo, TurnEndAction,
},
llm_client::{
ClientError, ConfigWarning, LlmClient, Request, RequestConfig, ResponseStream,
@@ -58,6 +62,9 @@ pub enum EngineError {
/// A durable-history observer rejected an item before it entered history.
#[error("History append failed: {0}")]
HistoryAppend(String),
/// A trusted host interceptor callback failed.
#[error(transparent)]
Interceptor(#[from] InterceptorFailure),
/// Tool terminalization lost its execution-attempt compare-and-set fence.
#[error("Tool execution attempt fence failed: {0}")]
ToolAttemptFence(String),
@@ -181,7 +188,7 @@ impl From<Result<EngineResult, EngineError>> for EngineRunExit {
/// Result of [`Engine::run`] or [`Engine::resume`].
///
/// Contains the `Locked` Engine (ready for subsequent runs) and the outcome.
pub struct EngineRunOutput<C: LlmClient, A = ()> {
pub struct EngineRunOutput<C: LlmClient, A: Send + Sync = ()> {
/// The Engine, now in Locked state.
pub engine: Engine<C, Locked, A>,
/// Outcome of the turn.
@@ -305,7 +312,7 @@ enum StreamCompletion {
Interrupted { reason: String },
}
pub struct Engine<C: LlmClient, S: EngineState = Mutable, A = ()> {
pub struct Engine<C: LlmClient, S: EngineState = Mutable, A: Send + Sync = ()> {
/// LLM client
client: C,
/// Retry policy for opening an LLM response stream.
@@ -322,7 +329,7 @@ pub struct Engine<C: LlmClient, S: EngineState = Mutable, A = ()> {
/// Tool server handle
tool_server: ToolServerHandle,
/// Interceptor for control-flow decisions
interceptor: Box<dyn Interceptor>,
interceptor: Box<dyn Interceptor<A>>,
/// System prompt
system_prompt: Option<String>,
/// History length at lock time (only meaningful in Locked state)
@@ -341,6 +348,11 @@ pub struct Engine<C: LlmClient, S: EngineState = Mutable, A = ()> {
/// `max_turns` is enforced against this run-scoped count rather than the
/// cumulative `turn_count` above.
active_run_turn_count: Option<usize>,
/// Identity retained across pause/yield and resume.
active_run_id: Option<InterceptorRunId>,
next_run_id: u64,
interceptor_invocation_count: usize,
last_run_exit_observer_failure: Option<InterceptorFailure>,
/// LlmCall count (per-Engine running counter, monotonic). Unlike
/// `turn_count` this never collapses retries.
llm_call_count: usize,
@@ -421,21 +433,57 @@ pub struct Engine<C: LlmClient, S: EngineState = Mutable, A = ()> {
_state: PhantomData<(S, A)>,
}
impl<C: LlmClient, S: EngineState, A> Engine<C, S, A> {
impl<C: LlmClient, S: EngineState, A: Send + Sync> Engine<C, S, A> {
fn start_logical_run(&mut self) {
self.active_run_turn_count = Some(0);
self.active_run_id = Some(InterceptorRunId(self.next_run_id));
self.next_run_id = self.next_run_id.wrapping_add(1).max(1);
self.interceptor_invocation_count = 0;
self.last_run_exit_observer_failure = None;
}
fn ensure_logical_run(&mut self) {
self.active_run_turn_count.get_or_insert(0);
if self.active_run_id.is_none() {
self.active_run_id = Some(InterceptorRunId(self.next_run_id));
self.next_run_id = self.next_run_id.wrapping_add(1).max(1);
self.interceptor_invocation_count = 0;
}
}
fn finish_logical_run(&mut self, result: &Result<EngineResult, EngineError>) {
if !matches!(
result,
Ok(EngineResult::Paused | EngineResult::Yielded) | Err(EngineError::PauseRequested)
) {
fn interceptor_invocation(
&mut self,
phase: InterceptorPhase,
turn_id: Option<usize>,
call_id: Option<InterceptorCallId>,
tool_call: usize,
) -> InterceptorInvocation {
let invocation = self.interceptor_invocation_count;
self.interceptor_invocation_count = self.interceptor_invocation_count.saturating_add(1);
InterceptorInvocation {
run_id: self
.active_run_id
.expect("logical run identity must exist before interception"),
turn_id: turn_id.map(|value| InterceptorTurnId(value as u64)),
call_id,
phase,
counters: InterceptorCounters {
invocation: InterceptorCounter::from_usize(invocation),
engine_turn: InterceptorCounter::from_usize(self.turn_count),
run_turn: InterceptorCounter::from_usize(
self.active_run_turn_count.unwrap_or_default(),
),
llm_call: InterceptorCounter::from_usize(self.llm_call_count),
tool_batch: InterceptorCounter::from_usize(self.tool_execution_batch_count),
tool_call: InterceptorCounter::from_usize(tool_call),
},
}
}
fn finish_logical_run(&mut self, exit: &EngineRunExit) {
if !matches!(exit, EngineRunExit::Paused | EngineRunExit::Yielded) {
self.active_run_turn_count = None;
self.active_run_id = None;
}
}
@@ -741,7 +789,7 @@ impl<C: LlmClient, S: EngineState, A> Engine<C, S, A> {
/// The interceptor governs approval, skip, pause, and abort decisions
/// at key points in the execution loop. If not set, the default
/// interceptor is used (all Continue / Finish).
pub fn set_interceptor(&mut self, interceptor: impl Interceptor + 'static) {
pub fn set_interceptor(&mut self, interceptor: impl Interceptor<A> + 'static) {
self.interceptor = Box::new(interceptor);
}
@@ -842,6 +890,10 @@ impl<C: LlmClient, S: EngineState, A> Engine<C, S, A> {
///
/// `Some` is retained only while Pause or Yield permits a later
/// [`resume`](Self::resume). Terminal outcomes return this to `None`.
pub fn last_run_exit_observer_failure(&self) -> Option<&InterceptorFailure> {
self.last_run_exit_observer_failure.as_ref()
}
pub fn active_run_turn_count(&self) -> Option<usize> {
self.active_run_turn_count
}
@@ -853,6 +905,13 @@ impl<C: LlmClient, S: EngineState, A> Engine<C, S, A> {
/// [`resume`](Self::resume) starts a fresh budget.
pub fn set_active_run_turn_count(&mut self, turn_count: Option<usize>) {
self.active_run_turn_count = turn_count;
if turn_count.is_none() {
self.active_run_id = None;
} else if self.active_run_id.is_none() {
self.active_run_id = Some(InterceptorRunId(self.next_run_id));
self.next_run_id = self.next_run_id.wrapping_add(1).max(1);
self.interceptor_invocation_count = 0;
}
}
/// Get the current LlmCall count (per-Engine running counter, never
@@ -1078,24 +1137,28 @@ impl<C: LlmClient, S: EngineState, A> Engine<C, S, A> {
request
}
/// Hooks: on_prompt_submit
///
async fn finalize_interruption<T>(
async fn finalize_run_exit(
&mut self,
result: Result<T, EngineError>,
) -> Result<T, EngineError> {
match result {
Ok(value) => Ok(value),
Err(err) => {
let reason = match &err {
EngineError::Aborted(reason) => reason.clone(),
EngineError::Cancelled => "Cancelled".to_string(),
_ => err.to_string(),
};
self.interceptor.on_abort(&reason).await;
Err(err)
}
history: &History<A>,
result: Result<EngineResult, EngineError>,
) -> EngineRunExit {
let exit = EngineRunExit::from(result);
let invocation = self.interceptor_invocation(InterceptorPhase::RunExit, None, None, 0);
self.last_run_exit_observer_failure = None;
if let Err(error) = self
.interceptor
.on_run_exit(RunExitContext {
invocation,
exit: &exit,
history: history.entries(),
})
.await
{
self.last_run_exit_observer_failure =
Some(InterceptorFailure::new(InterceptorPhase::RunExit, error));
}
self.finish_logical_run(&exit);
exit
}
/// Check for pending tool calls (for resuming from Pause)
@@ -1166,21 +1229,60 @@ impl<C: LlmClient, S: EngineState, A> Engine<C, S, A> {
// Phase 1: Apply pre_tool_call interceptor (determine skip/abort/synthetic result)
let mut approved_calls = Vec::new();
for (call_index, mut tool_call) in tool_calls.into_iter().enumerate() {
let expected_tool_use_id = tool_call.id.clone();
let context = ToolExecutionContext::new(&tool_call.id, &batch_id, call_index);
if let Some((meta, tool)) = self.tool_server.get_tool(&tool_call.name) {
let invocation = self.interceptor_invocation(
InterceptorPhase::PreToolCall,
Some(self.turn_count.saturating_sub(1)),
Some(InterceptorCallId::Tool(expected_tool_use_id.clone())),
call_index,
);
let mut info = ToolCallInfo {
invocation,
history: history.entries(),
call: tool_call.clone(),
meta,
tool,
context,
};
match self.interceptor.pre_tool_call(&mut info).await {
let pre_tool_action =
self.interceptor
.pre_tool_call(&mut info)
.await
.map_err(|error| {
EngineError::from(InterceptorFailure::new(
InterceptorPhase::PreToolCall,
error,
))
})?;
if info.call.id != expected_tool_use_id {
return Err(InterceptorFailure::new(
InterceptorPhase::PreToolCall,
InterceptorError::new(
InterceptorErrorCategory::ContractViolation,
"pre-tool interceptor changed immutable tool call identity",
),
)
.into());
}
match pre_tool_action {
PreToolAction::Continue => {}
PreToolAction::Skip => {
continue;
}
PreToolAction::SyntheticResult(result) => {
if result.tool_use_id != expected_tool_use_id {
return Err(InterceptorFailure::new(
InterceptorPhase::PreToolCall,
InterceptorError::new(
InterceptorErrorCategory::ContractViolation,
"synthetic tool result changed immutable tool call identity",
),
)
.into());
}
let tool_call = info.call;
let mut context = info.context;
context.call_id = tool_call.id.clone();
@@ -1287,20 +1389,31 @@ impl<C: LlmClient, S: EngineState, A> Engine<C, S, A> {
let mut terminal_call_ids = HashSet::new();
let mut pause_requested = false;
let mut pause_deadline = None;
let mut batch_error = None;
let mut locally_enqueued_cancel = false;
for result in synthetic_results {
self.finalize_and_commit_tool_result(
history,
annotate,
result,
None,
&call_info_map,
&mut attempt_fence,
&mut terminal_call_ids,
)
.await?;
if let Err(error) = self
.finalize_and_commit_tool_result(
history,
annotate,
result,
None,
&call_info_map,
&mut attempt_fence,
&mut terminal_call_ids,
)
.await
&& batch_error.is_none()
{
batch_error = Some(error);
}
}
let mut futures = futures;
if batch_error.is_some() && !futures.is_empty() {
let _ = self.cancel_tx.try_send(());
locally_enqueued_cancel = true;
}
while !futures.is_empty() {
tokio::select! {
// If cancellation and a completed result are both ready, drain
@@ -1310,7 +1423,7 @@ impl<C: LlmClient, S: EngineState, A> Engine<C, S, A> {
result = futures.next() => {
let (attempt_id, result) =
result.expect("non-empty FuturesUnordered returns a result");
self.finalize_and_commit_tool_result(
if let Err(error) = self.finalize_and_commit_tool_result(
history,
annotate,
result,
@@ -1318,7 +1431,15 @@ impl<C: LlmClient, S: EngineState, A> Engine<C, S, A> {
&call_info_map,
&mut attempt_fence,
&mut terminal_call_ids,
).await?;
).await {
if batch_error.is_none() {
batch_error = Some(error);
}
if !futures.is_empty() {
let _ = self.cancel_tx.try_send(());
locally_enqueued_cancel = true;
}
}
}
pause = self.pause_rx.recv(), if !pause_requested => {
if pause.is_some() {
@@ -1335,6 +1456,7 @@ impl<C: LlmClient, S: EngineState, A> Engine<C, S, A> {
_ = tokio::time::sleep_until(pause_deadline.unwrap_or_else(TokioInstant::now)), if pause_deadline.is_some() => {
pause_deadline = None;
let _ = self.cancel_tx.try_send(());
locally_enqueued_cancel = true;
}
cancel = self.cancel_rx.recv() => {
if cancel.is_some() {
@@ -1380,7 +1502,7 @@ impl<C: LlmClient, S: EngineState, A> Engine<C, S, A> {
result = futures.next() => {
let (attempt_id, result) =
result.expect("non-empty FuturesUnordered returns a result");
self.finalize_and_commit_tool_result(
if let Err(error) = self.finalize_and_commit_tool_result(
history,
annotate,
result,
@@ -1388,7 +1510,11 @@ impl<C: LlmClient, S: EngineState, A> Engine<C, S, A> {
&call_info_map,
&mut attempt_fence,
&mut terminal_call_ids,
).await?;
).await
&& batch_error.is_none()
{
batch_error = Some(error);
}
}
_ = tokio::time::sleep_until(deadline) => break,
}
@@ -1402,7 +1528,7 @@ impl<C: LlmClient, S: EngineState, A> Engine<C, S, A> {
if let Some(handle) = execution_handles.get(call_id) {
handle.force_close();
}
self.finalize_and_commit_tool_result(
if let Err(error) = self.finalize_and_commit_tool_result(
history,
annotate,
ToolResult::outcome_unknown(call_id),
@@ -1410,11 +1536,18 @@ impl<C: LlmClient, S: EngineState, A> Engine<C, S, A> {
&call_info_map,
&mut attempt_fence,
&mut terminal_call_ids,
).await?;
).await
&& batch_error.is_none()
{
batch_error = Some(error);
}
}
}
self.timeline.abort_current_block();
if let Some(error) = batch_error.take() {
return Err(error);
}
if pause_requested {
return Ok(ToolExecutionResult::Paused);
}
@@ -1423,6 +1556,16 @@ impl<C: LlmClient, S: EngineState, A> Engine<C, S, A> {
}
}
// A result-biased ready sibling can empty the batch before the local
// cancel signal is selected. Never let that current-batch signal leak
// into the next run or resume call.
if locally_enqueued_cancel {
let _ = self.cancel_rx.try_recv();
}
if let Some(error) = batch_error {
self.timeline.abort_current_block();
return Err(error);
}
Ok(if pause_requested {
ToolExecutionResult::Paused
} else {
@@ -1466,31 +1609,13 @@ impl<C: LlmClient, S: EngineState, A> Engine<C, S, A> {
}
let call_info = call_info_map.get(&tool_result.tool_use_id);
let mut abort_reason = None;
if let Some((tool_call, meta, tool, context)) = call_info {
let mut info = ToolResultInfo {
call: tool_call.clone(),
result: tool_result,
meta: meta.clone(),
tool: tool.clone(),
context: context.clone(),
};
match self.interceptor.post_tool_call(&mut info).await {
PostToolAction::Continue => {}
PostToolAction::Abort(reason) => {
abort_reason = Some(reason);
}
}
tool_result = info.result;
}
if tool_result.is_error && tool_result.disposition.is_success() {
tool_result.disposition = ToolResultDisposition::Error;
}
tool_result.is_error = !tool_result.disposition.is_success();
// Cap content only after post_tool_call so interceptors still observe
// the full payload and any content they inject is bounded too.
// Bound the terminal payload before committing it so the post-tool
// interceptor observes exactly the model-visible durable result.
if let (Some(limits), Some((tool_call, _, _, _)), Some(content)) = (
self.tool_output_limits.as_ref(),
call_info,
@@ -1543,9 +1668,38 @@ impl<C: LlmClient, S: EngineState, A> Engine<C, S, A> {
"Tool execution terminalized"
);
self.emit_tool_result(&tool_result);
if let Some(reason) = abort_reason {
return Err(EngineError::Aborted(reason));
if let Some((tool_call, meta, tool, context)) = call_info {
let invocation = self.interceptor_invocation(
InterceptorPhase::PostToolCall,
Some(self.turn_count.saturating_sub(1)),
Some(InterceptorCallId::Tool(tool_call.id.clone())),
context.call_index,
);
let info = ToolResultInfo {
invocation,
history: history.entries(),
call: tool_call.clone(),
result: tool_result,
meta: meta.clone(),
tool: tool.clone(),
context: context.clone(),
};
let post_tool_action =
self.interceptor
.post_tool_call(&info)
.await
.map_err(|error| {
EngineError::from(InterceptorFailure::new(
InterceptorPhase::PostToolCall,
error,
))
})?;
if let PostToolAction::Abort(reason) = post_tool_action {
return Err(EngineError::Aborted(reason));
}
}
Ok(true)
}
@@ -1608,11 +1762,25 @@ impl<C: LlmClient, S: EngineState, A> Engine<C, S, A> {
// These are committed *before* the per-request clone so they
// participate in the LLM request below and get persisted by
// the caller that owns durable history.
let pending_invocation = self.interceptor_invocation(
InterceptorPhase::PendingHistoryAppends,
Some(current_turn),
None,
0,
);
let pending = self
.interceptor
.pending_history_appends()
.pending_history_appends(PendingHistoryAppendsContext {
invocation: pending_invocation,
history: history.entries(),
})
.await
.map_err(EngineError::HistoryAppend)?;
.map_err(|error| {
EngineError::from(InterceptorFailure::new(
InterceptorPhase::PendingHistoryAppends,
error,
))
})?;
if !pending.is_empty() {
self.append_history_items(history, pending, annotate)?;
}
@@ -1679,7 +1847,27 @@ impl<C: LlmClient, S: EngineState, A> Engine<C, S, A> {
}
// Interceptor: pre_llm_request
match self.interceptor.pre_llm_request(&mut request_context).await {
let request_invocation = self.interceptor_invocation(
InterceptorPhase::PreLlmRequest,
Some(current_turn),
Some(InterceptorCallId::Llm(self.llm_call_count as u64)),
0,
);
let pre_request_action = self
.interceptor
.pre_llm_request(PreLlmRequestContext {
invocation: request_invocation,
items: &mut request_context,
history: history.entries(),
})
.await
.map_err(|error| {
EngineError::from(InterceptorFailure::new(
InterceptorPhase::PreLlmRequest,
error,
))
})?;
match pre_request_action {
PreRequestAction::Cancel(reason) => {
info!(reason = %reason, "Aborted by interceptor");
for cb in &self.turn_end_cbs {
@@ -1791,21 +1979,45 @@ impl<C: LlmClient, S: EngineState, A> Engine<C, S, A> {
let tool_calls = self.tool_call_collector.take_collected();
let assistant_items =
self.build_assistant_items(&reasoning_items, &text_blocks, &tool_calls);
let assistant_start = history.len();
self.append_history_items(history, assistant_items, annotate)?;
if tool_calls.is_empty() {
let turn_end_context = history.items_cloned();
match self.interceptor.on_turn_end(&turn_end_context).await {
TurnEndAction::Finish => {
return Ok(EngineResult::Finished);
}
TurnEndAction::ContinueWithMessages(additional) => {
self.append_history_items(history, additional, annotate)?;
let assistant_invocation = self.interceptor_invocation(
InterceptorPhase::AssistantTurnEnd,
Some(current_turn),
Some(InterceptorCallId::Llm(
self.llm_call_count.saturating_sub(1) as u64,
)),
0,
);
let assistant_turn_action = self
.interceptor
.on_assistant_turn_end(AssistantTurnEndContext {
invocation: assistant_invocation,
assistant_entries: &history.entries()[assistant_start..],
history: history.entries(),
tool_calls: &tool_calls,
})
.await
.map_err(|error| {
EngineError::from(InterceptorFailure::new(
InterceptorPhase::AssistantTurnEnd,
error,
))
})?;
match assistant_turn_action {
TurnEndAction::Finish if tool_calls.is_empty() => {
return Ok(EngineResult::Finished);
}
TurnEndAction::Finish => {}
TurnEndAction::ContinueWithMessages(additional) => {
self.append_history_items(history, additional, annotate)?;
if tool_calls.is_empty() {
continue;
}
TurnEndAction::Pause => {
return Ok(EngineResult::Paused);
}
}
TurnEndAction::Pause => {
return Ok(EngineResult::Paused);
}
}
@@ -2098,7 +2310,7 @@ impl<C: LlmClient, S: EngineState, A> Engine<C, S, A> {
}
}
impl<C: LlmClient, A> Engine<C, Mutable, A> {
impl<C: LlmClient, A: Send + Sync> Engine<C, Mutable, A> {
/// Create a new annotated Engine (in Mutable state).
pub fn new_annotated(client: C) -> Self {
let text_block_collector = TextBlockCollector::new();
@@ -2126,6 +2338,10 @@ impl<C: LlmClient, A> Engine<C, Mutable, A> {
locked_prefix_len: 0,
turn_count: 0,
active_run_turn_count: None,
active_run_id: None,
next_run_id: 1,
interceptor_invocation_count: 0,
last_run_exit_observer_failure: None,
llm_call_count: 0,
tool_execution_batch_count: 0,
max_turns: None,
@@ -2401,6 +2617,10 @@ impl<C: LlmClient, A> Engine<C, Mutable, A> {
locked_prefix_len,
turn_count: self.turn_count,
active_run_turn_count: self.active_run_turn_count,
active_run_id: self.active_run_id,
next_run_id: self.next_run_id,
interceptor_invocation_count: self.interceptor_invocation_count,
last_run_exit_observer_failure: self.last_run_exit_observer_failure,
llm_call_count: self.llm_call_count,
tool_execution_batch_count: self.tool_execution_batch_count,
max_turns: self.max_turns,
@@ -2477,7 +2697,7 @@ impl<C: LlmClient> Engine<C, Mutable, ()> {
}
}
impl<C: LlmClient, A> Engine<C, Locked, A> {
impl<C: LlmClient, A: Send + Sync> Engine<C, Locked, A> {
/// Execute a turn
///
/// Adds a new user message to history and sends a request to the LLM.
@@ -2488,9 +2708,10 @@ impl<C: LlmClient, A> Engine<C, Locked, A> {
user_input: impl Into<String>,
annotate: &mut impl FnMut(&Item) -> Result<A, String>,
) -> EngineRunExit {
self.run_result_with_annotation(history, user_input.into(), annotate)
.await
.into()
let result = self
.run_result_with_annotation(history, user_input.into(), annotate)
.await;
self.finalize_run_exit(history, result).await
}
async fn run_result_with_annotation(
@@ -2501,13 +2722,26 @@ impl<C: LlmClient, A> Engine<C, Locked, A> {
) -> Result<EngineResult, EngineError> {
// Supplying new user input abandons any paused/yielded logical run.
self.active_run_turn_count = None;
self.active_run_id = None;
self.start_logical_run();
let mut user_item = Item::user_message(user_input);
let extras = match self.interceptor.on_prompt_submit(&mut user_item).await {
PromptAction::Cancel(reason) => {
return self
.finalize_interruption(Err(EngineError::Aborted(reason)))
.await;
}
let invocation = self.interceptor_invocation(InterceptorPhase::PromptSubmit, None, None, 0);
let prompt_action = self
.interceptor
.on_prompt_submit(PromptSubmitContext {
invocation,
item: &mut user_item,
history: history.entries(),
})
.await
.map_err(|error| {
EngineError::from(InterceptorFailure::new(
InterceptorPhase::PromptSubmit,
error,
))
})?;
let extras = match prompt_action {
PromptAction::Cancel(reason) => return Err(EngineError::Aborted(reason)),
PromptAction::Continue => Vec::new(),
PromptAction::ContinueWith(items) => items,
};
@@ -2515,14 +2749,10 @@ impl<C: LlmClient, A> Engine<C, Locked, A> {
if !extras.is_empty() {
self.append_history_items(history, extras, annotate)?;
}
self.start_logical_run();
let result = match self.run_turn_loop(history, annotate).await {
match self.run_turn_loop(history, annotate).await {
Err(EngineError::PauseRequested) => Ok(EngineResult::Paused),
other => other,
};
let result = self.finalize_interruption(result).await;
self.finish_logical_run(&result);
result
}
}
/// Resume execution (from Paused state).
@@ -2531,9 +2761,8 @@ impl<C: LlmClient, A> Engine<C, Locked, A> {
history: &mut History<A>,
annotate: &mut impl FnMut(&Item) -> Result<A, String>,
) -> EngineRunExit {
self.resume_result_with_annotation(history, annotate)
.await
.into()
let result = self.resume_result_with_annotation(history, annotate).await;
self.finalize_run_exit(history, result).await
}
async fn resume_result_with_annotation(
@@ -2542,13 +2771,10 @@ impl<C: LlmClient, A> Engine<C, Locked, A> {
annotate: &mut impl FnMut(&Item) -> Result<A, String>,
) -> Result<EngineResult, EngineError> {
self.ensure_logical_run();
let result = match self.run_turn_loop(history, annotate).await {
match self.run_turn_loop(history, annotate).await {
Err(EngineError::PauseRequested) => Ok(EngineResult::Paused),
other => other,
};
let result = self.finalize_interruption(result).await;
self.finish_logical_run(&result);
result
}
}
/// Get the prefix length at lock time
@@ -2574,6 +2800,10 @@ impl<C: LlmClient, A> Engine<C, Locked, A> {
locked_prefix_len: 0,
turn_count: self.turn_count,
active_run_turn_count: self.active_run_turn_count,
active_run_id: self.active_run_id,
next_run_id: self.next_run_id,
interceptor_invocation_count: self.interceptor_invocation_count,
last_run_exit_observer_failure: self.last_run_exit_observer_failure,
llm_call_count: self.llm_call_count,
tool_execution_batch_count: self.tool_execution_batch_count,
max_turns: self.max_turns,
+250 -28
View File
@@ -9,8 +9,202 @@ use std::sync::Arc;
use async_trait::async_trait;
use crate::Item;
use crate::engine::EngineRunExit;
use crate::history::HistoryEntry;
use crate::tool::{Tool, ToolCall, ToolExecutionContext, ToolMeta, ToolResult};
// =============================================================================
// Typed lifecycle metadata and failures
// =============================================================================
/// Maximum UTF-8 byte length retained for interceptor diagnostics.
pub const MAX_INTERCEPTOR_DIAGNOSTIC_BYTES: usize = 1024;
/// Stable category for the source of an interceptor failure.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum InterceptorErrorCategory {
Policy,
Dependency,
ContractViolation,
Internal,
}
impl std::fmt::Display for InterceptorErrorCategory {
fn fmt(&self, formatter: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
formatter.write_str(match self {
Self::Policy => "policy",
Self::Dependency => "dependency",
Self::ContractViolation => "contract_violation",
Self::Internal => "internal",
})
}
}
/// A typed, bounded failure returned by an [`Interceptor`] implementation.
#[derive(Debug, Clone, PartialEq, Eq, thiserror::Error)]
#[error("{category}: {diagnostic}")]
pub struct InterceptorError {
category: InterceptorErrorCategory,
diagnostic: String,
}
impl InterceptorError {
pub fn new(category: InterceptorErrorCategory, diagnostic: impl Into<String>) -> Self {
let mut diagnostic = diagnostic.into();
if diagnostic.len() > MAX_INTERCEPTOR_DIAGNOSTIC_BYTES {
let mut end = MAX_INTERCEPTOR_DIAGNOSTIC_BYTES;
while !diagnostic.is_char_boundary(end) {
end -= 1;
}
diagnostic.truncate(end);
}
Self {
category,
diagnostic,
}
}
pub fn category(&self) -> InterceptorErrorCategory {
self.category
}
pub fn diagnostic(&self) -> &str {
&self.diagnostic
}
}
/// The lifecycle phase at which an interceptor callback executes.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Default)]
pub enum InterceptorPhase {
#[default]
PromptSubmit,
PendingHistoryAppends,
PreLlmRequest,
PreToolCall,
PostToolCall,
AssistantTurnEnd,
RunExit,
}
impl std::fmt::Display for InterceptorPhase {
fn fmt(&self, formatter: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
formatter.write_str(match self {
Self::PromptSubmit => "prompt_submit",
Self::PendingHistoryAppends => "pending_history_appends",
Self::PreLlmRequest => "pre_llm_request",
Self::PreToolCall => "pre_tool_call",
Self::PostToolCall => "post_tool_call",
Self::AssistantTurnEnd => "assistant_turn_end",
Self::RunExit => "run_exit",
})
}
}
#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash, Default)]
pub struct InterceptorRunId(pub u64);
#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
pub struct InterceptorTurnId(pub u64);
#[derive(Debug, Clone, PartialEq, Eq, Hash)]
pub enum InterceptorCallId {
Llm(u64),
Tool(String),
}
/// Saturating public counter used by interceptor contexts.
#[derive(Debug, Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Default)]
pub struct InterceptorCounter(u32);
impl InterceptorCounter {
pub fn from_usize(value: usize) -> Self {
Self(u32::try_from(value).unwrap_or(u32::MAX))
}
pub fn get(self) -> u32 {
self.0
}
}
#[derive(Debug, Clone, Copy, PartialEq, Eq, Default)]
pub struct InterceptorCounters {
pub invocation: InterceptorCounter,
pub engine_turn: InterceptorCounter,
pub run_turn: InterceptorCounter,
pub llm_call: InterceptorCounter,
pub tool_batch: InterceptorCounter,
pub tool_call: InterceptorCounter,
}
/// Identity, phase, and bounded counters common to every lifecycle callback.
#[derive(Debug, Clone, PartialEq, Eq, Default)]
pub struct InterceptorInvocation {
pub run_id: InterceptorRunId,
pub turn_id: Option<InterceptorTurnId>,
pub call_id: Option<InterceptorCallId>,
pub phase: InterceptorPhase,
pub counters: InterceptorCounters,
}
/// An interceptor failure bound to the exact Engine lifecycle phase that ran it.
#[derive(Debug, Clone, PartialEq, Eq, thiserror::Error)]
#[error("{phase} interceptor failed: {error}")]
pub struct InterceptorFailure {
phase: InterceptorPhase,
#[source]
error: InterceptorError,
}
impl InterceptorFailure {
pub(crate) fn new(phase: InterceptorPhase, error: InterceptorError) -> Self {
Self { phase, error }
}
pub fn phase(&self) -> InterceptorPhase {
self.phase
}
pub fn error(&self) -> &InterceptorError {
&self.error
}
}
pub type InterceptorResult<T> = Result<T, InterceptorError>;
// =============================================================================
// Lifecycle Contexts
// =============================================================================
pub struct PromptSubmitContext<'a, A = ()> {
pub invocation: InterceptorInvocation,
pub item: &'a mut Item,
pub history: &'a [HistoryEntry<A>],
}
pub struct PendingHistoryAppendsContext<'a, A = ()> {
pub invocation: InterceptorInvocation,
pub history: &'a [HistoryEntry<A>],
}
pub struct PreLlmRequestContext<'a, A = ()> {
pub invocation: InterceptorInvocation,
pub items: &'a mut Vec<Item>,
pub history: &'a [HistoryEntry<A>],
}
pub struct AssistantTurnEndContext<'a, A = ()> {
pub invocation: InterceptorInvocation,
pub assistant_entries: &'a [HistoryEntry<A>],
pub history: &'a [HistoryEntry<A>],
pub tool_calls: &'a [ToolCall],
}
pub struct RunExitContext<'a, A = ()> {
pub invocation: InterceptorInvocation,
pub exit: &'a EngineRunExit,
pub history: &'a [HistoryEntry<A>],
}
// =============================================================================
// Action Enums
// =============================================================================
@@ -86,9 +280,9 @@ pub enum PostToolAction {
/// Action at the end of a turn (when LLM produces no tool calls).
#[derive(Debug, Clone)]
pub enum TurnEndAction {
/// Turn is finished, return to caller.
/// Accept the Engine's natural next phase: execute tools, or finish when none exist.
Finish,
/// Continue with additional messages injected into history.
/// Commit additional messages, then continue through the natural next phase.
ContinueWithMessages(Vec<Item>),
/// Pause execution (can be resumed later).
Pause,
@@ -99,8 +293,9 @@ pub enum TurnEndAction {
// =============================================================================
/// Context for pre-tool-call decisions.
pub struct ToolCallInfo {
/// Tool call information (modifiable).
pub struct ToolCallInfo<'a, A = ()> {
pub invocation: InterceptorInvocation,
pub history: &'a [HistoryEntry<A>],
pub call: ToolCall,
/// Tool meta information.
pub meta: ToolMeta,
@@ -111,10 +306,11 @@ pub struct ToolCallInfo {
}
/// Context for post-tool-call decisions.
pub struct ToolResultInfo {
/// Original tool call.
pub struct ToolResultInfo<'a, A = ()> {
pub invocation: InterceptorInvocation,
pub history: &'a [HistoryEntry<A>],
pub call: ToolCall,
/// Tool execution result (modifiable).
/// Committed terminal tool execution result.
pub result: ToolResult,
/// Tool meta information.
pub meta: ToolMeta,
@@ -130,14 +326,22 @@ pub struct ToolResultInfo {
/// Intercepts the Engine execution loop at key decision points.
///
/// All methods have default implementations that let the Engine
/// proceed without intervention. Callers provide richer implementations for
/// approval flows, permission checks, etc.
/// Every lifecycle method is asynchronous and returns [`InterceptorResult`],
/// keeping implementation failure separate from the method's control-flow
/// action. The Engine reports a failure as a typed run interruption annotated
/// with the exact [`InterceptorPhase`] that failed.
///
/// All methods have default implementations that let the Engine proceed
/// without intervention. Callers provide richer implementations for approval
/// flows, permission checks, and other trusted host adaptation.
#[async_trait]
pub trait Interceptor: Send + Sync {
/// Called after receiving user input, before adding to history.
async fn on_prompt_submit(&self, _item: &mut Item) -> PromptAction {
PromptAction::Continue
pub trait Interceptor<A: Send + Sync = ()>: Send + Sync {
/// Called after receiving user input, before adding it to Engine history.
async fn on_prompt_submit(
&self,
_context: PromptSubmitContext<'_, A>,
) -> InterceptorResult<PromptAction> {
Ok(PromptAction::Continue)
}
/// Items that should be **committed to `engine.history`** just
@@ -158,7 +362,10 @@ pub trait Interceptor: Send + Sync {
/// reproducible per-request transformations (pruning, content
/// trimming, cache anchors) that depend only on the existing
/// history.
async fn pending_history_appends(&self) -> Result<Vec<Item>, String> {
async fn pending_history_appends(
&self,
_context: PendingHistoryAppendsContext<'_, A>,
) -> InterceptorResult<Vec<Item>> {
Ok(Vec::new())
}
@@ -170,27 +377,42 @@ pub trait Interceptor: Send + Sync {
/// If an interceptor derives a human/model-visible nudge from the current
/// request context, return [`PreRequestAction::ContinueWith`] so the Engine
/// commits it to history before the request is sent.
async fn pre_llm_request(&self, _context: &mut Vec<Item>) -> PreRequestAction {
PreRequestAction::Continue
async fn pre_llm_request(
&self,
_context: PreLlmRequestContext<'_, A>,
) -> InterceptorResult<PreRequestAction> {
Ok(PreRequestAction::Continue)
}
/// Called before each tool is executed.
async fn pre_tool_call(&self, _info: &mut ToolCallInfo) -> PreToolAction {
PreToolAction::Continue
async fn pre_tool_call(
&self,
_info: &mut ToolCallInfo<'_, A>,
) -> InterceptorResult<PreToolAction> {
Ok(PreToolAction::Continue)
}
/// Called after each tool completes.
async fn post_tool_call(&self, _info: &mut ToolResultInfo) -> PostToolAction {
PostToolAction::Continue
/// Called after each tool reaches one terminal result and that result is committed.
async fn post_tool_call(
&self,
_info: &ToolResultInfo<'_, A>,
) -> InterceptorResult<PostToolAction> {
Ok(PostToolAction::Continue)
}
/// Called when a turn ends with no tool calls.
async fn on_turn_end(&self, _history: &[Item]) -> TurnEndAction {
TurnEndAction::Finish
/// Called after every terminal assistant response is committed and before
/// the Engine decides whether to execute tools, continue, or finish.
async fn on_assistant_turn_end(
&self,
_context: AssistantTurnEndContext<'_, A>,
) -> InterceptorResult<TurnEndAction> {
Ok(TurnEndAction::Finish)
}
/// Called when execution is interrupted (abort or cancel).
async fn on_abort(&self, _reason: &str) {}
/// Called once for the terminal outcome of each public run or resume call.
async fn on_run_exit(&self, _context: RunExitContext<'_, A>) -> InterceptorResult<()> {
Ok(())
}
}
/// Default interceptor: no intervention. Engine proceeds through the loop
@@ -198,4 +420,4 @@ pub trait Interceptor: Send + Sync {
pub(crate) struct DefaultInterceptor;
#[async_trait]
impl Interceptor for DefaultInterceptor {}
impl<A: Send + Sync> Interceptor<A> for DefaultInterceptor {}
+7 -1
View File
@@ -26,7 +26,13 @@ pub use engine::{
};
pub use handler::ToolUseBlockStart;
pub use history::{History, HistoryEntry};
pub use interceptor::Interceptor;
pub use interceptor::{
AssistantTurnEndContext, Interceptor, InterceptorCallId, InterceptorCounter,
InterceptorCounters, InterceptorError, InterceptorErrorCategory, InterceptorFailure,
InterceptorInvocation, InterceptorPhase, InterceptorResult, InterceptorRunId,
InterceptorTurnId, MAX_INTERCEPTOR_DIAGNOSTIC_BYTES, PendingHistoryAppendsContext,
PreLlmRequestContext, PromptSubmitContext, RunExitContext,
};
pub use message::{ContentPart, Item, Message, Role};
pub use tool::{
ToolCall, ToolExecutionContext, ToolExecutionHandle, ToolExecutionPolicy,
+126
View File
@@ -1,8 +1,15 @@
mod common;
use agen::interceptor::{
AssistantTurnEndContext, Interceptor, InterceptorCallId, InterceptorInvocation,
InterceptorPhase, InterceptorResult, PendingHistoryAppendsContext, PreLlmRequestContext,
PreRequestAction, PromptAction, PromptSubmitContext, RunExitContext, TurnEndAction,
};
use agen::llm_client::event::{Event, ResponseStatus, StatusEvent};
use agen::{Engine, EngineError, History, HistoryEntry, Item, Role};
use async_trait::async_trait;
use common::MockLlmClient;
use std::sync::{Arc, Mutex};
fn completed_text_events(text: &str) -> Vec<Event> {
vec![
@@ -47,6 +54,125 @@ async fn run_preserves_item_annotations_without_projecting_them() {
assert_eq!(history.items_cloned().len(), 2);
}
#[derive(Clone)]
struct AnnotationObservingInterceptor {
observed: Arc<Mutex<Vec<(InterceptorInvocation, Vec<String>)>>>,
}
impl AnnotationObservingInterceptor {
fn record(&self, invocation: &InterceptorInvocation, history: &[HistoryEntry<String>]) {
self.observed.lock().unwrap().push((
invocation.clone(),
history
.iter()
.map(|entry| entry.annotation.clone())
.collect(),
));
}
}
#[async_trait]
impl Interceptor<String> for AnnotationObservingInterceptor {
async fn on_prompt_submit(
&self,
context: PromptSubmitContext<'_, String>,
) -> InterceptorResult<PromptAction> {
self.record(&context.invocation, context.history);
Ok(PromptAction::Continue)
}
async fn pending_history_appends(
&self,
context: PendingHistoryAppendsContext<'_, String>,
) -> InterceptorResult<Vec<Item>> {
self.record(&context.invocation, context.history);
Ok(Vec::new())
}
async fn pre_llm_request(
&self,
context: PreLlmRequestContext<'_, String>,
) -> InterceptorResult<PreRequestAction> {
self.record(&context.invocation, context.history);
Ok(PreRequestAction::Continue)
}
async fn on_assistant_turn_end(
&self,
context: AssistantTurnEndContext<'_, String>,
) -> InterceptorResult<TurnEndAction> {
assert_eq!(context.assistant_entries.len(), 1);
assert_eq!(context.assistant_entries[0].annotation, "2:assistant");
self.record(&context.invocation, context.history);
Ok(TurnEndAction::Finish)
}
async fn on_run_exit(&self, context: RunExitContext<'_, String>) -> InterceptorResult<()> {
self.record(&context.invocation, context.history);
Ok(())
}
}
#[tokio::test]
async fn interceptor_contexts_preserve_annotations_and_typed_lifecycle_identity() {
let client = MockLlmClient::new(completed_text_events("assistant reply"));
let mut engine = Engine::<_, agen::state::Mutable, String>::new_annotated(client);
let observed = Arc::new(Mutex::new(Vec::new()));
engine.set_interceptor(AnnotationObservingInterceptor {
observed: observed.clone(),
});
let mut history = History::<String>::new();
let mut next = 0usize;
let mut annotate = |item: &Item| {
next += 1;
let kind = if item.is_assistant_message() {
"assistant"
} else {
"user"
};
Ok(format!("{next}:{kind}"))
};
let output = engine
.run_with_annotation(&mut history, "hello", &mut annotate)
.await;
assert!(matches!(output.result, agen::EngineRunExit::Finished));
let observed = observed.lock().unwrap();
let phases: Vec<_> = observed
.iter()
.map(|(invocation, _)| invocation.phase)
.collect();
assert_eq!(
phases,
[
InterceptorPhase::PromptSubmit,
InterceptorPhase::PendingHistoryAppends,
InterceptorPhase::PreLlmRequest,
InterceptorPhase::AssistantTurnEnd,
InterceptorPhase::RunExit,
]
);
assert!(
observed
.iter()
.all(|(invocation, _)| invocation.run_id == observed[0].0.run_id)
);
assert_eq!(
observed
.iter()
.map(|(invocation, _)| invocation.counters.invocation.get())
.collect::<Vec<_>>(),
[0, 1, 2, 3, 4]
);
assert_eq!(observed[2].0.call_id, Some(InterceptorCallId::Llm(0)));
assert_eq!(observed[3].0.call_id, Some(InterceptorCallId::Llm(0)));
assert_eq!(observed[1].1, ["1:user"]);
assert_eq!(observed[2].1, ["1:user"]);
assert_eq!(observed[3].1, ["1:user", "2:assistant"]);
assert_eq!(observed[4].1, ["1:user", "2:assistant"]);
}
#[test]
fn append_failure_does_not_make_item_live() {
let client = MockLlmClient::new(vec![]);
+519 -10
View File
@@ -10,9 +10,16 @@ use std::sync::{Arc, Mutex};
use agen::Item;
use agen::interceptor::{
Interceptor, PreRequestAction, PreToolAction, ToolCallInfo, TurnEndAction,
AssistantTurnEndContext, Interceptor, InterceptorError, InterceptorErrorCategory,
InterceptorPhase as InterceptorPoint, InterceptorResult, MAX_INTERCEPTOR_DIAGNOSTIC_BYTES,
PendingHistoryAppendsContext, PostToolAction, PreLlmRequestContext, PreRequestAction,
PreToolAction, PromptAction, PromptSubmitContext, RunExitContext, ToolCallInfo, ToolResultInfo,
TurnEndAction,
};
use agen::llm_client::{
ClientError, LlmClient, Request, ResponseStream,
event::{Event, ResponseStatus, StatusEvent},
};
use agen::llm_client::event::{Event, ResponseStatus, StatusEvent};
use agen::tool::{Tool, ToolDefinition, ToolError, ToolMeta, ToolOutput};
use agen::{Engine, EngineError, EngineRunExit, History, RunInterruptionReason};
use async_trait::async_trait;
@@ -613,12 +620,15 @@ struct YieldOnce {
#[async_trait]
impl Interceptor for YieldOnce {
async fn pre_llm_request(&self, _context: &mut Vec<Item>) -> PreRequestAction {
if self.calls.fetch_add(1, Ordering::SeqCst) == 0 {
async fn pre_llm_request(
&self,
_context: PreLlmRequestContext<'_, ()>,
) -> InterceptorResult<PreRequestAction> {
Ok(if self.calls.fetch_add(1, Ordering::SeqCst) == 0 {
PreRequestAction::Yield
} else {
PreRequestAction::Continue
}
})
}
}
@@ -628,12 +638,15 @@ struct PauseToolOnce {
#[async_trait]
impl Interceptor for PauseToolOnce {
async fn pre_tool_call(&self, _info: &mut ToolCallInfo) -> PreToolAction {
if self.calls.fetch_add(1, Ordering::SeqCst) == 0 {
async fn pre_tool_call(
&self,
_info: &mut ToolCallInfo<'_, ()>,
) -> InterceptorResult<PreToolAction> {
Ok(if self.calls.fetch_add(1, Ordering::SeqCst) == 0 {
PreToolAction::Pause
} else {
PreToolAction::Continue
}
})
}
}
@@ -643,13 +656,509 @@ struct ContinueTurnOnce {
#[async_trait]
impl Interceptor for ContinueTurnOnce {
async fn on_turn_end(&self, _history: &[Item]) -> TurnEndAction {
if self.calls.fetch_add(1, Ordering::SeqCst) == 0 {
async fn on_assistant_turn_end(
&self,
_context: AssistantTurnEndContext<'_, ()>,
) -> InterceptorResult<TurnEndAction> {
Ok(if self.calls.fetch_add(1, Ordering::SeqCst) == 0 {
TurnEndAction::ContinueWithMessages(vec![Item::system_message("continue")])
} else {
TurnEndAction::Finish
})
}
}
#[derive(Debug, Clone)]
struct FailingLifecycleInterceptor {
failure: InterceptorPoint,
calls: Arc<Mutex<Vec<InterceptorPoint>>>,
}
impl FailingLifecycleInterceptor {
fn new(failure: InterceptorPoint) -> Self {
Self {
failure,
calls: Arc::new(Mutex::new(Vec::new())),
}
}
fn record<T>(&self, point: InterceptorPoint, action: T) -> InterceptorResult<T> {
self.calls.lock().unwrap().push(point);
if self.failure == point {
Err(InterceptorError::new(
InterceptorErrorCategory::Policy,
format!("{point} rejected"),
))
} else {
Ok(action)
}
}
fn calls(&self) -> Vec<InterceptorPoint> {
self.calls.lock().unwrap().clone()
}
}
#[async_trait]
impl Interceptor for FailingLifecycleInterceptor {
async fn on_prompt_submit(
&self,
_context: PromptSubmitContext<'_, ()>,
) -> InterceptorResult<PromptAction> {
tokio::task::yield_now().await;
self.record(InterceptorPoint::PromptSubmit, PromptAction::Continue)
}
async fn pending_history_appends(
&self,
_context: PendingHistoryAppendsContext<'_, ()>,
) -> InterceptorResult<Vec<Item>> {
tokio::task::yield_now().await;
self.record(InterceptorPoint::PendingHistoryAppends, Vec::new())
}
async fn pre_llm_request(
&self,
_context: PreLlmRequestContext<'_, ()>,
) -> InterceptorResult<PreRequestAction> {
tokio::task::yield_now().await;
self.record(InterceptorPoint::PreLlmRequest, PreRequestAction::Continue)
}
async fn pre_tool_call(
&self,
_info: &mut ToolCallInfo<'_, ()>,
) -> InterceptorResult<PreToolAction> {
tokio::task::yield_now().await;
self.record(InterceptorPoint::PreToolCall, PreToolAction::Continue)
}
async fn post_tool_call(
&self,
_info: &ToolResultInfo<'_, ()>,
) -> InterceptorResult<PostToolAction> {
tokio::task::yield_now().await;
self.record(InterceptorPoint::PostToolCall, PostToolAction::Continue)
}
async fn on_assistant_turn_end(
&self,
context: AssistantTurnEndContext<'_, ()>,
) -> InterceptorResult<TurnEndAction> {
tokio::task::yield_now().await;
assert!(context.history.ends_with(context.assistant_entries));
if !context.tool_calls.is_empty() {
assert_eq!(
context
.assistant_entries
.iter()
.filter(|entry| matches!(&entry.item, Item::ToolCall { .. }))
.count(),
context.tool_calls.len()
);
}
self.record(InterceptorPoint::AssistantTurnEnd, TurnEndAction::Finish)
}
async fn on_run_exit(&self, _context: RunExitContext<'_, ()>) -> InterceptorResult<()> {
tokio::task::yield_now().await;
self.record(InterceptorPoint::RunExit, ())
}
}
fn expected_interceptor_calls(failure: InterceptorPoint) -> Vec<InterceptorPoint> {
use InterceptorPoint as Point;
let mut calls = match failure {
Point::PromptSubmit => vec![Point::PromptSubmit],
Point::PendingHistoryAppends => {
vec![Point::PromptSubmit, Point::PendingHistoryAppends]
}
Point::PreLlmRequest => vec![
Point::PromptSubmit,
Point::PendingHistoryAppends,
Point::PreLlmRequest,
],
Point::PreToolCall => vec![
Point::PromptSubmit,
Point::PendingHistoryAppends,
Point::PreLlmRequest,
Point::AssistantTurnEnd,
Point::PreToolCall,
],
Point::PostToolCall => vec![
Point::PromptSubmit,
Point::PendingHistoryAppends,
Point::PreLlmRequest,
Point::AssistantTurnEnd,
Point::PreToolCall,
Point::PostToolCall,
],
Point::AssistantTurnEnd => vec![
Point::PromptSubmit,
Point::PendingHistoryAppends,
Point::PreLlmRequest,
Point::AssistantTurnEnd,
],
Point::RunExit => vec![
Point::PromptSubmit,
Point::PendingHistoryAppends,
Point::PreLlmRequest,
Point::AssistantTurnEnd,
],
};
calls.push(Point::RunExit);
calls
}
#[tokio::test]
async fn interceptor_failures_are_typed_and_terminal_observer_preserves_original_exit() {
use InterceptorPoint as Point;
for failure_point in [
Point::PromptSubmit,
Point::PendingHistoryAppends,
Point::PreLlmRequest,
Point::PreToolCall,
Point::PostToolCall,
Point::AssistantTurnEnd,
Point::RunExit,
] {
let interceptor = FailingLifecycleInterceptor::new(failure_point);
let needs_tool = matches!(failure_point, Point::PreToolCall | Point::PostToolCall);
let events = if needs_tool {
vec![
Event::tool_use_start(0, "call-1", "count_tool"),
Event::tool_input_delta(0, "{}"),
Event::tool_use_stop(0),
Event::Status(StatusEvent {
status: ResponseStatus::Completed,
}),
]
} else {
completed_text_events()
};
let mut engine = Engine::new(MockLlmClient::new(events));
engine.register_tool(CountingTool::new("count_tool").definition());
engine.set_interceptor(interceptor.clone());
let mut history = History::new();
let mut engine = engine.lock(&history);
let exit = engine.run(&mut history, "test").await;
let failure = if failure_point == Point::RunExit {
assert!(matches!(exit, EngineRunExit::Finished));
engine
.last_run_exit_observer_failure()
.expect("terminal observer diagnostic should be retained")
} else {
let EngineRunExit::Interrupted(RunInterruptionReason::Unexpected(
EngineError::Interceptor(failure),
)) = &exit
else {
panic!("expected typed interceptor interruption at {failure_point}, got {exit:?}");
};
failure
};
assert_eq!(failure.phase(), failure_point);
assert_eq!(
failure.error().diagnostic(),
format!("{failure_point} rejected")
);
assert_eq!(
interceptor.calls(),
expected_interceptor_calls(failure_point)
);
if failure_point == Point::PostToolCall {
assert!(
history
.items()
.any(|item| matches!(item, Item::ToolResult { .. })),
"post-tool failure must not precede terminal output commit"
);
}
}
}
#[test]
fn interceptor_error_keeps_typed_category_and_bounded_utf8_diagnostic() {
let error = InterceptorError::new(
InterceptorErrorCategory::Dependency,
"".repeat(MAX_INTERCEPTOR_DIAGNOSTIC_BYTES),
);
assert_eq!(error.category(), InterceptorErrorCategory::Dependency);
assert!(error.diagnostic().len() <= MAX_INTERCEPTOR_DIAGNOSTIC_BYTES);
assert!(
error
.diagnostic()
.is_char_boundary(error.diagnostic().len())
);
}
struct FailingRunExitObserver {
pause: bool,
}
#[async_trait]
impl Interceptor for FailingRunExitObserver {
async fn on_assistant_turn_end(
&self,
_context: AssistantTurnEndContext<'_, ()>,
) -> InterceptorResult<TurnEndAction> {
Ok(if self.pause {
TurnEndAction::Pause
} else {
TurnEndAction::Finish
})
}
async fn on_run_exit(&self, _context: RunExitContext<'_, ()>) -> InterceptorResult<()> {
Err(InterceptorError::new(
InterceptorErrorCategory::Dependency,
"terminal audit unavailable",
))
}
}
#[tokio::test]
async fn terminal_observer_failure_preserves_paused_and_interrupted_exits() {
let mut paused_engine = Engine::new(MockLlmClient::new(completed_text_events()));
paused_engine.set_interceptor(FailingRunExitObserver { pause: true });
let mut paused_history = History::new();
let mut paused_engine = paused_engine.lock(&paused_history);
assert!(matches!(
paused_engine.run(&mut paused_history, "pause").await,
EngineRunExit::Paused
));
assert_eq!(
paused_engine
.last_run_exit_observer_failure()
.expect("paused observer diagnostic")
.error()
.category(),
InterceptorErrorCategory::Dependency
);
let mut interrupted_engine = Engine::new(MockLlmClient::new(completed_text_events()));
interrupted_engine.set_max_turns(Some(0));
interrupted_engine.set_interceptor(FailingRunExitObserver { pause: false });
let mut interrupted_history = History::new();
let mut interrupted_engine = interrupted_engine.lock(&interrupted_history);
assert!(matches!(
interrupted_engine
.run(&mut interrupted_history, "limit")
.await,
EngineRunExit::Interrupted(RunInterruptionReason::LimitReached)
));
assert_eq!(
interrupted_engine
.last_run_exit_observer_failure()
.expect("interrupted observer diagnostic")
.phase(),
InterceptorPoint::RunExit
);
}
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
enum TerminalMode {
Finish,
PauseOnce,
Yield,
}
#[derive(Debug, Clone)]
struct RecordingTerminalInterceptor {
mode: TerminalMode,
assistant_turns: Arc<AtomicUsize>,
exits: Arc<Mutex<Vec<&'static str>>>,
}
impl RecordingTerminalInterceptor {
fn new(mode: TerminalMode) -> Self {
Self {
mode,
assistant_turns: Arc::new(AtomicUsize::new(0)),
exits: Arc::new(Mutex::new(Vec::new())),
}
}
fn exits(&self) -> Vec<&'static str> {
self.exits.lock().unwrap().clone()
}
}
#[async_trait]
impl Interceptor for RecordingTerminalInterceptor {
async fn pre_llm_request(
&self,
_context: PreLlmRequestContext<'_, ()>,
) -> InterceptorResult<PreRequestAction> {
Ok(if self.mode == TerminalMode::Yield {
PreRequestAction::Yield
} else {
PreRequestAction::Continue
})
}
async fn on_assistant_turn_end(
&self,
context: AssistantTurnEndContext<'_, ()>,
) -> InterceptorResult<TurnEndAction> {
assert!(!context.assistant_entries.is_empty());
assert!(
context.history.ends_with(context.assistant_entries),
"assistant-turn callback must observe committed terminal items"
);
let turn = self.assistant_turns.fetch_add(1, Ordering::SeqCst);
Ok(if self.mode == TerminalMode::PauseOnce && turn == 0 {
TurnEndAction::Pause
} else {
TurnEndAction::Finish
})
}
async fn on_run_exit(&self, context: RunExitContext<'_, ()>) -> InterceptorResult<()> {
let kind = match context.exit {
EngineRunExit::Finished => "finished",
EngineRunExit::Paused => "paused",
EngineRunExit::Yielded => "yielded",
EngineRunExit::Interrupted(RunInterruptionReason::LimitReached) => "limit",
EngineRunExit::Interrupted(RunInterruptionReason::ContextWindowExceeded) => "context",
EngineRunExit::Interrupted(RunInterruptionReason::Cancelled) => "cancelled",
EngineRunExit::Interrupted(RunInterruptionReason::Unexpected(_)) => "unexpected",
};
self.exits.lock().unwrap().push(kind);
Ok(())
}
}
#[derive(Clone)]
struct ContextWindowClient;
#[async_trait]
impl LlmClient for ContextWindowClient {
async fn stream(&self, _request: Request) -> Result<ResponseStream, ClientError> {
Err(ClientError::ContextWindowExceeded)
}
fn clone_boxed(&self) -> Box<dyn LlmClient> {
Box::new(self.clone())
}
}
#[tokio::test]
async fn terminal_observer_runs_once_for_every_exit_and_interruption_kind() {
let finished = RecordingTerminalInterceptor::new(TerminalMode::Finish);
let mut engine = Engine::new(MockLlmClient::new(completed_text_events()));
engine.set_interceptor(finished.clone());
let mut history = History::new();
assert!(matches!(
engine.lock(&history).run(&mut history, "finish").await,
EngineRunExit::Finished
));
assert_eq!(finished.exits(), ["finished"]);
let yielded = RecordingTerminalInterceptor::new(TerminalMode::Yield);
let mut engine = Engine::new(MockLlmClient::new(completed_text_events()));
engine.set_interceptor(yielded.clone());
let mut history = History::new();
assert!(matches!(
engine.lock(&history).run(&mut history, "yield").await,
EngineRunExit::Yielded
));
assert_eq!(yielded.exits(), ["yielded"]);
let limited = RecordingTerminalInterceptor::new(TerminalMode::Finish);
let mut engine = Engine::new(MockLlmClient::new(completed_text_events()));
engine.set_max_turns(Some(0));
engine.set_interceptor(limited.clone());
let mut history = History::new();
assert!(matches!(
engine.lock(&history).run(&mut history, "limit").await,
EngineRunExit::Interrupted(RunInterruptionReason::LimitReached)
));
assert_eq!(limited.exits(), ["limit"]);
let cancelled = RecordingTerminalInterceptor::new(TerminalMode::Finish);
let mut engine = Engine::new(MockLlmClient::new(completed_text_events()));
engine.set_interceptor(cancelled.clone());
engine.cancel();
let mut history = History::new();
assert!(matches!(
engine.lock(&history).run(&mut history, "cancel").await,
EngineRunExit::Interrupted(RunInterruptionReason::Cancelled)
));
assert_eq!(cancelled.exits(), ["cancelled"]);
let context = RecordingTerminalInterceptor::new(TerminalMode::Finish);
let mut engine = Engine::new(ContextWindowClient);
engine.set_interceptor(context.clone());
let mut history = History::new();
assert!(matches!(
engine.lock(&history).run(&mut history, "context").await,
EngineRunExit::Interrupted(RunInterruptionReason::ContextWindowExceeded)
));
assert_eq!(context.exits(), ["context"]);
let unexpected = FailingLifecycleInterceptor::new(InterceptorPoint::PromptSubmit);
let mut engine = Engine::new(MockLlmClient::new(completed_text_events()));
engine.set_interceptor(unexpected.clone());
let mut history = History::new();
assert!(matches!(
engine.lock(&history).run(&mut history, "fail").await,
EngineRunExit::Interrupted(RunInterruptionReason::Unexpected(EngineError::Interceptor(
_
)))
));
assert_eq!(
unexpected
.calls()
.iter()
.filter(|point| **point == InterceptorPoint::RunExit)
.count(),
1
);
}
#[tokio::test]
async fn terminal_observer_does_not_duplicate_on_resume() {
let interceptor = RecordingTerminalInterceptor::new(TerminalMode::PauseOnce);
let first_response = vec![
Event::tool_use_start(0, "call-1", "count_tool"),
Event::tool_input_delta(0, "{}"),
Event::tool_use_stop(0),
Event::Status(StatusEvent {
status: ResponseStatus::Completed,
}),
];
let client = MockLlmClient::with_responses(vec![first_response, completed_text_events()]);
let tool = CountingTool::new("count_tool");
let mut engine = Engine::new(client);
engine.register_tool(tool.definition());
engine.set_interceptor(interceptor.clone());
let mut history = History::new();
let mut engine = engine.lock(&history);
assert!(matches!(
engine.run(&mut history, "pause").await,
EngineRunExit::Paused
));
assert_eq!(interceptor.exits(), ["paused"]);
assert_eq!(
tool.call_count(),
0,
"pause must retain the pending tool phase"
);
assert!(matches!(
engine.resume(&mut history).await,
EngineRunExit::Finished
));
assert_eq!(interceptor.exits(), ["paused", "finished"]);
assert_eq!(
tool.call_count(),
1,
"resume must execute the retained tool once"
);
}
#[tokio::test]
+240 -40
View File
@@ -6,13 +6,18 @@ use std::sync::atomic::{AtomicUsize, Ordering};
use std::sync::{Arc, Mutex};
use std::time::{Duration, Instant};
use agen::interceptor::{Interceptor, PostToolAction, PreToolAction, ToolCallInfo, ToolResultInfo};
use agen::interceptor::{
Interceptor, InterceptorError, InterceptorErrorCategory, InterceptorPhase, InterceptorResult,
PostToolAction, PreToolAction, ToolCallInfo, ToolResultInfo,
};
use agen::llm_client::event::{Event, ResponseStatus, StatusEvent};
use agen::tool::{
Tool, ToolDefinition, ToolError, ToolExecutionContext, ToolMeta, ToolOutput, ToolResult,
ToolResultDisposition,
};
use agen::{Engine, History, Item, ToolExecutionPolicy};
use agen::{
Engine, EngineError, EngineRunExit, History, Item, RunInterruptionReason, ToolExecutionPolicy,
};
use async_trait::async_trait;
mod common;
@@ -905,24 +910,30 @@ async fn test_tool_execution_context_for_skipped_and_synthetic_paths() {
#[async_trait]
impl Interceptor for ContextPolicy {
async fn pre_tool_call(&self, info: &mut ToolCallInfo) -> PreToolAction {
async fn pre_tool_call(
&self,
info: &mut ToolCallInfo<'_, ()>,
) -> InterceptorResult<PreToolAction> {
self.pre_contexts.lock().unwrap().push(info.context.clone());
match info.call.name.as_str() {
Ok(match info.call.name.as_str() {
"skip_tool" => PreToolAction::Skip,
"synthetic_tool" => PreToolAction::SyntheticResult(ToolResult::from_output(
&info.call.id,
ToolOutput::from("synthetic result".to_string()),
)),
_ => PreToolAction::Continue,
}
})
}
async fn post_tool_call(&self, info: &mut ToolResultInfo) -> PostToolAction {
async fn post_tool_call(
&self,
info: &ToolResultInfo<'_, ()>,
) -> InterceptorResult<PostToolAction> {
self.post_contexts
.lock()
.unwrap()
.push(info.context.clone());
PostToolAction::Continue
Ok(PostToolAction::Continue)
}
}
@@ -994,12 +1005,15 @@ async fn test_before_tool_call_skip() {
#[async_trait]
impl Interceptor for BlockingPolicy {
async fn pre_tool_call(&self, info: &mut ToolCallInfo) -> PreToolAction {
if info.call.name == "blocked_tool" {
async fn pre_tool_call(
&self,
info: &mut ToolCallInfo<'_, ()>,
) -> InterceptorResult<PreToolAction> {
Ok(if info.call.name == "blocked_tool" {
PreToolAction::Skip
} else {
PreToolAction::Continue
}
})
}
}
@@ -1021,9 +1035,9 @@ async fn test_before_tool_call_skip() {
);
}
/// Hook: post_tool_call - verify that results can be modified
/// Hook: post_tool_call - verify that the committed terminal result is observed.
#[tokio::test]
async fn test_post_tool_call_modification() {
async fn test_post_tool_call_observes_committed_result() {
// Prepare responses for multiple requests
let client = MockLlmClient::with_responses(vec![
// First request: tool call
@@ -1074,40 +1088,51 @@ async fn test_post_tool_call_modification() {
engine.register_tool(simple_tool_definition());
// Policy to modify results
struct ModifyingPolicy {
modified_content: Arc<std::sync::Mutex<Option<String>>>,
// Policy to observe the committed terminal result.
struct ObservingPolicy {
observed_content: Arc<std::sync::Mutex<Option<String>>>,
}
#[async_trait]
impl Interceptor for ModifyingPolicy {
async fn post_tool_call(&self, info: &mut ToolResultInfo) -> PostToolAction {
info.result.summary = format!("[Modified] {}", info.result.summary);
*self.modified_content.lock().unwrap() = Some(info.result.summary.clone());
PostToolAction::Continue
impl Interceptor for ObservingPolicy {
async fn post_tool_call(
&self,
info: &ToolResultInfo<'_, ()>,
) -> InterceptorResult<PostToolAction> {
assert_eq!(info.invocation.phase, InterceptorPhase::PostToolCall);
assert_eq!(
info.invocation.call_id,
Some(agen::InterceptorCallId::Tool(info.call.id.clone()))
);
assert!(matches!(
info.history.last().map(|entry| &entry.item),
Some(Item::ToolResult { call_id, .. }) if call_id == &info.call.id
));
*self.observed_content.lock().unwrap() = Some(info.result.summary.clone());
Ok(PostToolAction::Continue)
}
}
let modified_content = Arc::new(std::sync::Mutex::new(None));
engine.set_interceptor(ModifyingPolicy {
modified_content: modified_content.clone(),
let observed_content = Arc::new(std::sync::Mutex::new(None));
engine.set_interceptor(ObservingPolicy {
observed_content: observed_content.clone(),
});
// Mutable::run consumes self, returns (Locked, EngineResult)
let result = engine.run(&mut history, "Test modification").await;
let result = engine.run(&mut history, "Test observation").await;
assert!(
matches!(result.result, agen::EngineRunExit::Finished),
"Engine should complete"
);
// Verify hook was called and content was modified
let content = modified_content.lock().unwrap().clone();
assert!(content.is_some(), "Hook should have been called");
assert!(
content.unwrap().contains("[Modified]"),
"Result should be modified"
);
// Verify the interceptor observed the exact committed result.
let observed = observed_content.lock().unwrap().clone();
assert_eq!(observed.as_deref(), Some("Original Result"));
assert!(history.items().any(|item| matches!(
item,
Item::ToolResult { summary, .. } if summary == "Original Result"
)));
}
/// Hook: pre_tool_call synthetic result - skipped tool gets an error result in history.
@@ -1143,11 +1168,14 @@ async fn test_before_tool_call_synthetic_result_committed() {
#[async_trait]
impl Interceptor for SyntheticPolicy {
async fn pre_tool_call(&self, info: &mut ToolCallInfo) -> PreToolAction {
PreToolAction::SyntheticResult(ToolResult::error(
async fn pre_tool_call(
&self,
info: &mut ToolCallInfo<'_, ()>,
) -> InterceptorResult<PreToolAction> {
Ok(PreToolAction::SyntheticResult(ToolResult::error(
info.call.id.clone(),
"permission denied",
))
)))
}
}
@@ -1167,6 +1195,80 @@ async fn test_before_tool_call_synthetic_result_committed() {
)));
}
#[derive(Clone, Copy)]
enum InvalidIdentityMode {
ContinuedCall,
SyntheticResult,
}
struct InvalidIdentityPolicy(InvalidIdentityMode);
#[async_trait]
impl Interceptor for InvalidIdentityPolicy {
async fn pre_tool_call(
&self,
info: &mut ToolCallInfo<'_, ()>,
) -> InterceptorResult<PreToolAction> {
assert_eq!(info.invocation.phase, InterceptorPhase::PreToolCall);
assert_eq!(
info.invocation.call_id,
Some(agen::InterceptorCallId::Tool("call_1".to_string()))
);
assert!(matches!(
info.history.last().map(|entry| &entry.item),
Some(Item::ToolCall { call_id, .. }) if call_id == "call_1"
));
Ok(match self.0 {
InvalidIdentityMode::ContinuedCall => {
info.call.id = "different-call".to_string();
PreToolAction::Continue
}
InvalidIdentityMode::SyntheticResult => PreToolAction::SyntheticResult(
ToolResult::error("different-call", "invalid synthetic result"),
),
})
}
}
#[tokio::test]
async fn interceptor_cannot_change_tool_call_identity() {
for mode in [
InvalidIdentityMode::ContinuedCall,
InvalidIdentityMode::SyntheticResult,
] {
let client = MockLlmClient::new(vec![
Event::tool_use_start(0, "call_1", "echo"),
Event::tool_input_delta(0, r#"{}"#),
Event::tool_use_stop(0),
Event::Status(StatusEvent {
status: ResponseStatus::Completed,
}),
]);
let mut engine = Engine::new(client);
engine.register_tool(SlowTool::new("echo", 1).definition());
engine.set_interceptor(InvalidIdentityPolicy(mode));
let mut history = History::new();
let result = engine.run(&mut history, "identity").await;
let EngineRunExit::Interrupted(RunInterruptionReason::Unexpected(
EngineError::Interceptor(failure),
)) = result.result
else {
panic!("invalid tool identity must interrupt with a typed failure");
};
assert_eq!(failure.phase(), InterceptorPhase::PreToolCall);
assert_eq!(
failure.error().category(),
InterceptorErrorCategory::ContractViolation
);
assert!(
!history
.items()
.any(|item| matches!(item, Item::ToolResult { .. }))
);
}
}
#[tokio::test]
async fn post_tool_abort_commits_confirmed_result_before_stopping_run() {
let client = MockLlmClient::new(vec![
@@ -1181,16 +1283,24 @@ async fn post_tool_abort_commits_confirmed_result_before_stopping_run() {
let tool = SlowTool::new("confirmed", 1);
engine.register_tool(tool.definition());
struct AbortAfterResult;
let observed = Arc::new(Mutex::new(Vec::<&'static str>::new()));
struct AbortAfterResult {
lifecycle: Arc<Mutex<Vec<&'static str>>>,
}
#[async_trait]
impl Interceptor for AbortAfterResult {
async fn post_tool_call(&self, _info: &mut ToolResultInfo) -> PostToolAction {
PostToolAction::Abort("policy stopped the run".to_string())
async fn post_tool_call(
&self,
_info: &ToolResultInfo<'_, ()>,
) -> InterceptorResult<PostToolAction> {
self.lifecycle.lock().unwrap().push("post_tool_call");
Ok(PostToolAction::Abort("policy stopped the run".to_string()))
}
}
engine.set_interceptor(AbortAfterResult);
engine.set_interceptor(AbortAfterResult {
lifecycle: observed.clone(),
});
let observed = Arc::new(Mutex::new(Vec::<&'static str>::new()));
let published = observed.clone();
engine.on_tool_result(move |_| published.lock().unwrap().push("published"));
let committed = observed.clone();
@@ -1210,7 +1320,7 @@ async fn post_tool_abort_commits_confirmed_result_before_stopping_run() {
assert_eq!(tool.call_count(), 1);
assert_eq!(
observed.lock().unwrap().as_slice(),
["committed", "published", "run-returned"]
["committed", "published", "post_tool_call", "run-returned"]
);
assert!(matches!(
output.result,
@@ -1239,3 +1349,93 @@ async fn post_tool_abort_commits_confirmed_result_before_stopping_run() {
} if call_id == "call_confirmed"
)));
}
#[derive(Clone, Copy)]
enum PostToolStopMode {
Abort,
Failure,
}
struct StopFirstParallelResult(PostToolStopMode);
#[async_trait]
impl Interceptor for StopFirstParallelResult {
async fn post_tool_call(
&self,
info: &ToolResultInfo<'_, ()>,
) -> InterceptorResult<PostToolAction> {
if info.call.id != "call_fast" {
return Ok(PostToolAction::Continue);
}
tokio::time::sleep(Duration::from_millis(5)).await;
match self.0 {
PostToolStopMode::Abort => Ok(PostToolAction::Abort("stop parallel batch".to_string())),
PostToolStopMode::Failure => Err(InterceptorError::new(
InterceptorErrorCategory::Policy,
"reject parallel batch",
)),
}
}
}
#[tokio::test]
async fn post_tool_stop_terminalizes_started_parallel_siblings_before_returning() {
for mode in [PostToolStopMode::Abort, PostToolStopMode::Failure] {
let first_response = vec![
Event::tool_use_start(0, "call_fast", "fast"),
Event::tool_input_delta(0, r#"{}"#),
Event::tool_use_stop(0),
Event::tool_use_start(1, "call_ready", "ready"),
Event::tool_input_delta(1, r#"{}"#),
Event::tool_use_stop(1),
Event::Status(StatusEvent {
status: ResponseStatus::Completed,
}),
];
let second_response = vec![
Event::text_block_start(0),
Event::text_delta(0, "next run completed"),
Event::text_block_stop(0, None),
Event::Status(StatusEvent {
status: ResponseStatus::Completed,
}),
];
let client = MockLlmClient::with_responses(vec![first_response, second_response]);
let mut engine = Engine::new(client);
engine.register_tool(SlowTool::new("fast", 0).definition());
engine.register_tool(SlowTool::new("ready", 1).definition());
engine.set_interceptor(StopFirstParallelResult(mode));
let mut history = History::new();
let output = engine.run(&mut history, "parallel stop").await;
match mode {
PostToolStopMode::Abort => assert!(matches!(
output.result,
EngineRunExit::Interrupted(RunInterruptionReason::Unexpected(
EngineError::Aborted(ref reason)
)) if reason == "stop parallel batch"
)),
PostToolStopMode::Failure => assert!(matches!(
output.result,
EngineRunExit::Interrupted(RunInterruptionReason::Unexpected(
EngineError::Interceptor(ref failure)
)) if failure.phase() == InterceptorPhase::PostToolCall
)),
}
let terminal_ids: Vec<_> = history
.iter()
.filter_map(|entry| match &entry.item {
Item::ToolResult { call_id, .. } => Some(call_id.as_str()),
_ => None,
})
.collect();
assert_eq!(terminal_ids.len(), 2);
assert!(terminal_ids.contains(&"call_fast"));
assert!(terminal_ids.contains(&"call_ready"));
let mut engine = output.engine;
let next = engine.run(&mut history, "next run").await;
assert!(matches!(next, EngineRunExit::Finished));
}
}
+167 -78
View File
@@ -18,10 +18,11 @@ use crate::model::{AuthRef, ModelManifest, ReasoningControl};
use crate::plugin::PluginConfig;
use crate::{
CompactionConfig, EngineManifest, FeatureConfig, FeatureFlagConfig, FileUploadLimits,
McpConfig, McpEnvValue, McpStdioCwdPolicy, MemoryConfig, MemoryFeatureConfig,
MergeRequestFeatureConfig, ScopeConfig, SessionConfig, SkillsConfig, TicketFeatureConfig,
ToolOutputLimits, ToolPermissionConfig, ToolPermissionRule, WebConfig, WorkerFeatureConfig,
WorkerManifest, WorkerMeta,
McpConfig, McpEnvValue, McpStdioCwdPolicy, MemoryConsolidationProfileConfig,
MemoryExtractionProfileConfig, MemoryFeatureProfileConfig, MemoryResidentProfileConfig,
MergeRequestFeatureConfig, ResolvedMemoryFeatureConfig, ScopeConfig, SessionConfig,
SkillsConfig, TicketFeatureConfig, ToolOutputLimits, ToolPermissionConfig, ToolPermissionRule,
WebConfig, WorkerFeatureConfig, WorkerManifest, WorkerMeta,
};
/// Partial-form Worker manifest. Every field is optional; one or more
@@ -67,9 +68,6 @@ pub struct WorkerManifestConfig {
/// First-class web tool opt-in. See [`WebConfig`].
#[serde(default)]
pub web: Option<WebConfig>,
/// Memory subsystem opt-in. See [`MemoryConfig`].
#[serde(default)]
pub memory: Option<MemoryConfig>,
/// External Agent Skills directories. See [`crate::SkillsConfig`].
#[serde(default)]
pub skills: Option<SkillsConfig>,
@@ -193,18 +191,86 @@ impl From<WorkerFeatureConfigPartial> for WorkerFeatureConfig {
}
#[derive(Debug, Clone, Default, Serialize, Deserialize)]
#[serde(deny_unknown_fields)]
pub struct MemoryFeatureConfigPartial {
#[serde(default)]
pub enabled: Option<bool>,
#[serde(default)]
pub staging: Option<bool>,
pub staging_tools: Option<bool>,
#[serde(default)]
pub resident: Option<MemoryResidentProfileConfigPartial>,
#[serde(default)]
pub extraction: Option<MemoryExtractionProfileConfigPartial>,
#[serde(default)]
pub consolidation: Option<MemoryConsolidationProfileConfigPartial>,
}
#[derive(Debug, Clone, Default, Serialize, Deserialize)]
#[serde(deny_unknown_fields)]
pub struct MemoryResidentProfileConfigPartial {
#[serde(default)]
pub inject_summary: Option<bool>,
}
#[derive(Debug, Clone, Default, Serialize, Deserialize)]
#[serde(deny_unknown_fields)]
pub struct MemoryExtractionProfileConfigPartial {
#[serde(default)]
pub enabled: Option<bool>,
#[serde(default)]
pub model: Option<ModelManifest>,
#[serde(default)]
pub threshold: Option<u64>,
#[serde(default)]
pub worker_max_turns: Option<u32>,
}
#[derive(Debug, Clone, Default, Serialize, Deserialize)]
#[serde(deny_unknown_fields)]
pub struct MemoryConsolidationProfileConfigPartial {
#[serde(default)]
pub request_enabled: Option<bool>,
}
impl MemoryFeatureConfigPartial {
fn merge(self, other: Self) -> Self {
Self {
enabled: other.enabled.or(self.enabled),
staging: other.staging.or(self.staging),
staging_tools: other.staging_tools.or(self.staging_tools),
resident: merge_option(
self.resident,
other.resident,
MemoryResidentProfileConfigPartial::merge,
),
extraction: merge_option(
self.extraction,
other.extraction,
MemoryExtractionProfileConfigPartial::merge,
),
consolidation: merge_option(
self.consolidation,
other.consolidation,
MemoryConsolidationProfileConfigPartial::merge,
),
}
}
}
impl MemoryResidentProfileConfigPartial {
fn merge(self, other: Self) -> Self {
Self {
inject_summary: other.inject_summary.or(self.inject_summary),
}
}
}
impl MemoryExtractionProfileConfigPartial {
fn merge(self, other: Self) -> Self {
Self {
enabled: other.enabled.or(self.enabled),
model: other.model.or(self.model),
threshold: other.threshold.or(self.threshold),
worker_max_turns: other.worker_max_turns.or(self.worker_max_turns),
}
}
}
@@ -253,13 +319,21 @@ impl MergeRequestFeatureConfigPartial {
}
}
impl MemoryConsolidationProfileConfigPartial {
fn merge(self, other: Self) -> Self {
Self {
request_enabled: other.request_enabled.or(self.request_enabled),
}
}
}
impl From<FeatureConfigPartial> for FeatureConfig {
fn from(value: FeatureConfigPartial) -> Self {
Self {
task: value.task.map(FeatureFlagConfig::from).unwrap_or_default(),
memory: value
.memory
.map(MemoryFeatureConfig::from)
.map(ResolvedMemoryFeatureConfig::from)
.unwrap_or_default(),
web: value.web.map(FeatureFlagConfig::from).unwrap_or_default(),
image: value.image.map(FeatureFlagConfig::from).unwrap_or_default(),
@@ -329,20 +403,52 @@ impl From<WorkerFeatureConfig> for WorkerFeatureConfigPartial {
}
}
impl From<MemoryFeatureConfigPartial> for MemoryFeatureConfig {
impl From<MemoryFeatureConfigPartial> for ResolvedMemoryFeatureConfig {
fn from(value: MemoryFeatureConfigPartial) -> Self {
let resident = value.resident.unwrap_or_default();
let extraction = value.extraction.unwrap_or_default();
let consolidation = value.consolidation.unwrap_or_default();
Self {
enabled: value.enabled.unwrap_or_default(),
staging: value.staging.unwrap_or_default(),
profile: MemoryFeatureProfileConfig {
enabled: value.enabled.unwrap_or_default(),
staging_tools: value.staging_tools.unwrap_or_default(),
resident: MemoryResidentProfileConfig {
inject_summary: resident.inject_summary.unwrap_or(true),
},
extraction: MemoryExtractionProfileConfig {
enabled: extraction.enabled.unwrap_or(true),
model: extraction.model,
threshold: extraction.threshold.or(Some(50_000)),
worker_max_turns: extraction
.worker_max_turns
.or(defaults::MEMORY_EXTRACT_WORKER_MAX_TURNS),
},
consolidation: MemoryConsolidationProfileConfig {
request_enabled: consolidation.request_enabled.unwrap_or(true),
},
},
workspace_settings: None,
}
}
}
impl From<MemoryFeatureConfig> for MemoryFeatureConfigPartial {
fn from(value: MemoryFeatureConfig) -> Self {
impl From<ResolvedMemoryFeatureConfig> for MemoryFeatureConfigPartial {
fn from(value: ResolvedMemoryFeatureConfig) -> Self {
Self {
enabled: Some(value.enabled),
staging: Some(value.staging),
enabled: Some(value.profile.enabled),
staging_tools: Some(value.profile.staging_tools),
resident: Some(MemoryResidentProfileConfigPartial {
inject_summary: Some(value.profile.resident.inject_summary),
}),
extraction: Some(MemoryExtractionProfileConfigPartial {
enabled: Some(value.profile.extraction.enabled),
model: value.profile.extraction.model,
threshold: value.profile.extraction.threshold,
worker_max_turns: value.profile.extraction.worker_max_turns,
}),
consolidation: Some(MemoryConsolidationProfileConfigPartial {
request_enabled: Some(value.profile.consolidation.request_enabled),
}),
}
}
}
@@ -543,13 +649,9 @@ pub(crate) fn reject_removed_manifest_fields(s: &str) -> Result<(), toml::de::Er
(removed; use compaction.prune_protected_tokens)",
));
}
if value
.get("memory")
.and_then(toml::Value::as_table)
.is_some_and(|table| table.contains_key("extract_worker_max_input_tokens"))
{
if value.get("memory").is_some() {
return Err(toml::de::Error::custom(
"unknown field in manifest: memory.extract_worker_max_input_tokens (removed)",
"unknown field in manifest: memory (removed; configure feature.memory)",
));
}
if value
@@ -633,11 +735,6 @@ impl WorkerManifestConfig {
for rule in &mut self.delegation_scope.deny {
rule.target = join_if_relative(base, &rule.target);
}
if let Some(ref mut memory) = self.memory
&& let Some(ref mut root) = memory.workspace_root
{
*root = join_if_relative(base, root);
}
if let Some(ref mut compaction) = self.compaction
&& let Some(ref mut cp) = compaction.model
{
@@ -682,7 +779,6 @@ impl WorkerManifestConfig {
CompactionConfigPartial::merge,
),
web: merge_option(self.web, upper.web, WebConfig::merge),
memory: merge_option(self.memory, upper.memory, MemoryConfig::merge),
skills: merge_option(self.skills, upper.skills, SkillsConfig::merge),
}
}
@@ -754,32 +850,6 @@ impl crate::WebFetchConfig {
}
}
impl MemoryConfig {
fn merge(self, upper: Self) -> Self {
Self {
workspace_root: upper.workspace_root.or(self.workspace_root),
query_result_limit: upper.query_result_limit.or(self.query_result_limit),
query_excerpt_lines: upper.query_excerpt_lines.or(self.query_excerpt_lines),
inject_summary: upper.inject_summary.or(self.inject_summary),
workspace_id: upper.workspace_id.or(self.workspace_id),
settings_revision: upper.settings_revision.or(self.settings_revision),
language: upper.language.or(self.language),
extract_model: upper.extract_model.or(self.extract_model),
extract_threshold: upper.extract_threshold.or(self.extract_threshold),
extract_worker_max_turns: upper
.extract_worker_max_turns
.or(self.extract_worker_max_turns),
consolidation_model: upper.consolidation_model.or(self.consolidation_model),
consolidation_threshold_files: upper
.consolidation_threshold_files
.or(self.consolidation_threshold_files),
consolidation_threshold_bytes: upper
.consolidation_threshold_bytes
.or(self.consolidation_threshold_bytes),
}
}
}
impl WorkerMetaConfig {
fn merge(self, upper: Self) -> Self {
Self {
@@ -1223,7 +1293,6 @@ impl TryFrom<WorkerManifestConfig> for WorkerManifest {
mcp: cfg.mcp,
compaction,
web: cfg.web,
memory: cfg.memory,
skills: cfg.skills,
profile: None,
})
@@ -1271,7 +1340,6 @@ mod tests {
session: None,
compaction: None,
web: None,
memory: None,
skills: None,
}
}
@@ -1846,29 +1914,50 @@ prune_protected_turns = 3
}
#[test]
fn from_toml_rejects_removed_extract_worker_max_input_tokens_field() {
let bad = r#"
[memory]
extract_worker_max_input_tokens = 30000
"#;
let err = WorkerManifestConfig::from_toml(bad).unwrap_err();
assert!(
err.to_string()
.contains("memory.extract_worker_max_input_tokens"),
"unexpected error: {err}"
);
fn from_toml_accepts_memory_extraction_settings_only_under_feature_memory() {
let cfg = WorkerManifestConfig::from_toml(
r#"
[feature.memory]
enabled = true
staging_tools = false
[feature.memory.resident]
inject_summary = false
[feature.memory.extraction]
enabled = true
threshold = 42000
worker_max_turns = 2
[feature.memory.consolidation]
request_enabled = false
"#,
)
.unwrap();
let memory = cfg.feature.memory.unwrap();
assert_eq!(memory.enabled, Some(true));
assert_eq!(memory.staging_tools, Some(false));
assert_eq!(memory.resident.unwrap().inject_summary, Some(false));
assert_eq!(memory.consolidation.unwrap().request_enabled, Some(false));
let extraction = memory.extraction.unwrap();
assert_eq!(extraction.enabled, Some(true));
assert_eq!(extraction.threshold, Some(42_000));
assert_eq!(extraction.worker_max_turns, Some(2));
}
#[test]
fn from_toml_accepts_extract_worker_max_turns() {
let cfg = WorkerManifestConfig::from_toml(
fn from_toml_rejects_legacy_top_level_memory_authority() {
let err = WorkerManifestConfig::from_toml(
r#"
[memory]
extract_worker_max_turns = 2
"#,
)
.unwrap();
assert_eq!(cfg.memory.unwrap().extract_worker_max_turns, Some(2));
.unwrap_err();
assert!(
err.to_string().contains("memory"),
"unexpected error: {err}"
);
}
#[test]
@@ -1948,7 +2037,7 @@ worker_max_turns = 7
fn feature_flags_default_disabled_in_resolved_manifest() {
let manifest: WorkerManifest = minimal_valid().try_into().unwrap();
assert!(!manifest.feature.task.enabled);
assert!(!manifest.feature.memory.enabled);
assert!(!manifest.feature.memory.profile.enabled);
assert!(!manifest.feature.web.enabled);
assert!(!manifest.feature.sub_worker.enabled);
assert!(!manifest.feature.objective.enabled);
@@ -2025,8 +2114,8 @@ enabled = false
}
);
assert!(!manifest.feature.orchestration.enabled);
assert!(!manifest.feature.memory.enabled);
assert!(!manifest.feature.memory.staging);
assert!(!manifest.feature.memory.profile.enabled);
assert!(!manifest.feature.memory.profile.staging_tools);
assert!(!manifest.feature.objective.enabled);
}
@@ -2074,7 +2163,7 @@ readiness_check = true
enabled = true
[feature.memory]
staging = true
staging_tools = true
[feature.manage_workdir]
enabled = true
@@ -2111,8 +2200,8 @@ enabled = true
})
.try_into()
.unwrap();
assert!(manifest.feature.memory.enabled);
assert!(manifest.feature.memory.staging);
assert!(manifest.feature.memory.profile.enabled);
assert!(manifest.feature.memory.profile.staging_tools);
assert!(manifest.feature.manage_workdir.enabled);
assert!(manifest.feature.ticket.enabled);
assert!(!manifest.feature.ticket.authoring);
+1 -1
View File
@@ -93,5 +93,5 @@ pub const COMPACT_RESULT_CONTEXT_MAX_TOKENS: u64 = 60_000;
pub const COMPACT_DEFAULT_REFERENCE_COUNT: usize = 5;
/// Optional maximum extract-worker tool-loop depth. `None` means unlimited.
/// See [`crate::MemoryConfig::extract_worker_max_turns`].
/// See [`crate::MemoryExtractionProfileConfig::worker_max_turns`].
pub const MEMORY_EXTRACT_WORKER_MAX_TURNS: Option<u32> = Some(8);
+506 -147
View File
@@ -47,6 +47,7 @@ use serde::{Deserialize, Serialize};
/// part of the manifest — it is the process's `std::env::current_dir()`
/// at construction time.
#[derive(Debug, Clone, Serialize, Deserialize)]
#[serde(deny_unknown_fields)]
pub struct WorkerManifest {
pub worker: WorkerMeta,
pub model: ModelManifest,
@@ -80,11 +81,6 @@ pub struct WorkerManifest {
pub mcp: McpConfig,
#[serde(default)]
pub compaction: Option<CompactionConfig>,
/// Memory subsystem configuration. Presence of `[memory]` configures memory
/// storage, extraction, consolidation, and resident injection, but memory
/// tools are surfaced only when `[feature.memory].enabled = true`.
#[serde(default)]
pub memory: Option<MemoryConfig>,
/// First-class web tools configuration. Network access remains fail-closed
/// under this config; WebSearch/WebFetch schemas are surfaced only when
/// `[feature.web].enabled = true`.
@@ -109,12 +105,12 @@ pub struct WorkerManifest {
/// profile/config data only: they do not carry runtime Worker names, sockets,
/// sessions, secrets, or resolved host state. Tool registration still applies
/// the normal scope, host-authority, backend, memory, and network checks.
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
pub struct FeatureConfig {
#[serde(default)]
pub task: FeatureFlagConfig,
#[serde(default)]
pub memory: MemoryFeatureConfig,
pub memory: ResolvedMemoryFeatureConfig,
#[serde(default)]
pub web: FeatureFlagConfig,
#[serde(default)]
@@ -147,7 +143,7 @@ impl Default for FeatureConfig {
fn default() -> Self {
Self {
task: FeatureFlagConfig::disabled(),
memory: MemoryFeatureConfig::disabled(),
memory: ResolvedMemoryFeatureConfig::default(),
web: FeatureFlagConfig::disabled(),
image: FeatureFlagConfig::disabled(),
sub_worker: FeatureFlagConfig::disabled(),
@@ -222,34 +218,139 @@ const fn default_true() -> bool {
true
}
#[derive(Debug, Clone, Copy, Serialize, Deserialize, PartialEq, Eq)]
pub struct MemoryFeatureConfig {
#[serde(default)]
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
#[serde(default, deny_unknown_fields)]
pub struct MemoryFeatureProfileConfig {
pub enabled: bool,
/// Exposes Memory staging queue tools in addition to normal Memory CRUD/query tools.
#[serde(default)]
pub staging: bool,
pub staging_tools: bool,
pub resident: MemoryResidentProfileConfig,
pub extraction: MemoryExtractionProfileConfig,
pub consolidation: MemoryConsolidationProfileConfig,
}
impl MemoryFeatureConfig {
pub const fn disabled() -> Self {
Self {
enabled: false,
staging: false,
}
impl MemoryFeatureProfileConfig {
pub fn disabled() -> Self {
Self::default()
}
pub const fn enabled() -> Self {
pub fn enabled() -> Self {
Self {
enabled: true,
staging: false,
..Self::default()
}
}
}
impl Default for MemoryFeatureConfig {
impl Default for MemoryFeatureProfileConfig {
fn default() -> Self {
Self::disabled()
Self {
enabled: false,
staging_tools: false,
resident: MemoryResidentProfileConfig::default(),
extraction: MemoryExtractionProfileConfig::default(),
consolidation: MemoryConsolidationProfileConfig::default(),
}
}
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
#[serde(default, deny_unknown_fields)]
pub struct MemoryResidentProfileConfig {
pub inject_summary: bool,
}
impl Default for MemoryResidentProfileConfig {
fn default() -> Self {
Self {
inject_summary: true,
}
}
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
#[serde(default, deny_unknown_fields)]
pub struct MemoryExtractionProfileConfig {
pub enabled: bool,
pub model: Option<ModelManifest>,
pub threshold: Option<u64>,
pub worker_max_turns: Option<u32>,
}
impl Default for MemoryExtractionProfileConfig {
fn default() -> Self {
Self {
enabled: true,
model: None,
threshold: Some(50_000),
worker_max_turns: defaults::MEMORY_EXTRACT_WORKER_MAX_TURNS,
}
}
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
#[serde(default, deny_unknown_fields)]
pub struct MemoryConsolidationProfileConfig {
pub request_enabled: bool,
}
impl Default for MemoryConsolidationProfileConfig {
fn default() -> Self {
Self {
request_enabled: true,
}
}
}
/// Immutable Memory execution configuration persisted in a resolved Worker Manifest.
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Default)]
#[serde(default, deny_unknown_fields)]
pub struct ResolvedMemoryFeatureConfig {
pub profile: MemoryFeatureProfileConfig,
#[serde(skip_serializing_if = "Option::is_none")]
pub workspace_settings: Option<WorkspaceMemorySettingsSnapshot>,
}
impl ResolvedMemoryFeatureConfig {
pub fn enabled(&self) -> bool {
self.profile.enabled
}
pub fn bind_workspace_settings(
&mut self,
settings: WorkspaceMemorySettingsSnapshot,
) -> Result<(), &'static str> {
if !self.profile.enabled {
if self.workspace_settings.is_some() {
return Err("disabled Memory feature must not carry Workspace settings");
}
return Ok(());
}
if self.workspace_settings.is_some() {
return Err("memory Workspace settings are already bound");
}
self.workspace_settings = Some(settings);
Ok(())
}
pub fn workspace_settings(&self) -> Option<WorkspaceMemorySettingsSnapshot> {
self.workspace_settings.clone()
}
pub fn validate_execution(&self) -> Result<(), &'static str> {
if self.profile.enabled && self.workspace_settings.is_none() {
return Err("enabled Memory feature requires trusted Workspace settings");
}
if !self.profile.enabled && self.workspace_settings.is_some() {
return Err("disabled Memory feature must not carry Workspace settings");
}
if let Some(settings) = &self.workspace_settings
&& (settings.settings_revision == 0
|| !is_normalized_workspace_memory_language(&settings.language))
{
return Err("Memory Workspace settings snapshot metadata is invalid");
}
Ok(())
}
}
@@ -484,98 +585,6 @@ pub struct WorkspaceMemorySettingsSnapshot {
pub language: String,
}
/// Memory subsystem configuration. Presence in the manifest enables
/// memory; `workspace_root` pins the memory workspace explicitly. When it
/// is absent, memory resolution searches upward from the Worker's pwd for a
/// `.yoi/memory` marker rather than treating `.yoi` project records alone
/// as a memory root.
///
/// All fields are `Option`; defaults are applied at the consumer
/// (`.unwrap_or(defaults::...)`). This keeps cascade `merge` simple
/// (`upper.x.or(self.x)`) without a separate partial/resolved split.
#[derive(Debug, Clone, Default, Serialize, Deserialize)]
pub struct MemoryConfig {
/// Override for the memory workspace root. When `None`, consumers resolve
/// the root from their default path and ancestor `.yoi/memory` markers.
/// When set, must be an absolute path.
#[serde(default)]
pub workspace_root: Option<PathBuf>,
/// Maximum number of records returned by `MemoryQuery` /
/// `MemoryQuery` per call. `None` ⇒ tool default (20).
#[serde(default)]
pub query_result_limit: Option<usize>,
/// Lines of context before and after each match in query excerpts.
/// Ignored when the request omits `query`. `None` ⇒ tool default (3).
#[serde(default)]
pub query_excerpt_lines: Option<usize>,
/// Whether the body of `memory/summary.md` is exposed in the resident
/// system-prompt section. `None` ⇒ enabled.
#[serde(default)]
pub inject_summary: Option<bool>,
/// Workspace that owns the bound Memory settings revision.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub workspace_id: Option<String>,
/// Monotonic revision of the bound Workspace Memory settings.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub settings_revision: Option<u64>,
/// Language from the bound Workspace Memory settings revision.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub language: Option<String>,
/// Optional model for the extract worker. When `None`,
/// the main engine model is cloned via `clone_boxed()`. Lightweight
/// reasoning-capable models (Haiku / 4o-mini / Flash class) are
/// recommended.
#[serde(default)]
pub extract_model: Option<ModelManifest>,
/// Cumulative input-token threshold (since the last extract pointer)
/// that triggers an extract run. `None` disables the extract trigger
/// entirely; memory tools and resident injection still work, only
/// the auto-extract trigger is dormant.
#[serde(default)]
pub extract_threshold: Option<u64>,
/// Optional maximum extract-worker tool-loop depth. `None` leaves
/// the worker unlimited; the default bounds runaway short-context
/// loops. Falls through to
/// [`defaults::MEMORY_EXTRACT_WORKER_MAX_TURNS`] when unset.
#[serde(default)]
pub extract_worker_max_turns: Option<u32>,
/// Optional model for the consolidation worker. When
/// `None`, the main engine model is cloned via `clone_boxed()`.
/// Reasoning-class models are recommended.
#[serde(default)]
pub consolidation_model: Option<ModelManifest>,
/// Consolidation trigger: file-count threshold of `_staging/`. The
/// consolidation run fires when the staging directory has at least
/// this many entries. Either threshold reaching its limit fires
/// consolidation (logical OR). `None` for both thresholds ⇒
/// consolidation disabled.
#[serde(default)]
pub consolidation_threshold_files: Option<usize>,
/// Consolidation trigger: byte-size threshold across all `_staging/`
/// entries. Either threshold reaching its limit fires consolidation.
/// `None` for both thresholds ⇒ consolidation disabled.
#[serde(default)]
pub consolidation_threshold_bytes: Option<u64>,
}
impl MemoryConfig {
/// Replace any untrusted manifest values with a trusted Workspace snapshot.
pub fn bind_workspace_settings(&mut self, snapshot: &WorkspaceMemorySettingsSnapshot) {
self.workspace_id = Some(snapshot.workspace_id.clone());
self.settings_revision = Some(snapshot.settings_revision);
self.language = Some(snapshot.language.clone());
}
/// Return the complete bound Workspace settings snapshot, if every field is present.
pub fn workspace_settings(&self) -> Option<WorkspaceMemorySettingsSnapshot> {
Some(WorkspaceMemorySettingsSnapshot {
workspace_id: self.workspace_id.clone()?,
settings_revision: self.settings_revision?,
language: self.language.clone()?,
})
}
}
/// Worker metadata.
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct WorkerMeta {
@@ -931,6 +940,12 @@ impl Default for CompactionConfig {
}
impl WorkerManifest {
pub fn requires_persisted_execution_snapshot(&self) -> bool {
self.profile.is_some()
|| self.plugins.has_resolved_plan()
|| self.feature.memory.workspace_settings.is_some()
}
/// Parse a manifest from a TOML string.
pub fn from_toml(s: &str) -> Result<Self, toml::de::Error> {
config::reject_removed_manifest_fields(s)?;
@@ -941,6 +956,212 @@ impl WorkerManifest {
}
}
#[derive(Debug, Default, Deserialize)]
#[serde(default, deny_unknown_fields)]
struct LegacyMemoryFeatureConfig {
enabled: bool,
staging: bool,
}
#[derive(Debug, Default, Deserialize)]
#[serde(default, deny_unknown_fields)]
struct LegacyMemoryConfig {
#[serde(rename = "workspace_root")]
_workspace_root: Option<PathBuf>,
#[serde(rename = "query_result_limit")]
_query_result_limit: Option<usize>,
#[serde(rename = "query_excerpt_lines")]
_query_excerpt_lines: Option<usize>,
inject_summary: Option<bool>,
workspace_id: Option<String>,
settings_revision: Option<u64>,
language: Option<String>,
extract_model: Option<ModelManifest>,
extract_threshold: Option<u64>,
extract_worker_max_turns: Option<u32>,
consolidation_model: Option<ModelManifest>,
consolidation_threshold_files: Option<usize>,
consolidation_threshold_bytes: Option<u64>,
}
const RESOLVED_MANIFEST_SNAPSHOT_SCHEMA_VERSION: u64 = 2;
/// Serialize a resolved Worker Manifest for durable Worker-specific storage.
pub fn write_persisted_worker_manifest_snapshot(
manifest: &WorkerManifest,
) -> Result<serde_json::Value, serde_json::Error> {
Ok(serde_json::json!({
"schema_version": RESOLVED_MANIFEST_SNAPSHOT_SCHEMA_VERSION,
"manifest": serde_json::to_value(manifest)?,
}))
}
/// Read a durable resolved Worker Manifest through the versioned compatibility
/// boundary. Runtime code must not deserialize persisted snapshots directly.
pub fn read_persisted_worker_manifest_snapshot(
snapshot: serde_json::Value,
) -> Result<WorkerManifest, serde_json::Error> {
let object = snapshot.as_object().ok_or_else(|| {
serde_json::Error::io(std::io::Error::new(
std::io::ErrorKind::InvalidData,
"resolved Worker manifest snapshot must be an object",
))
})?;
if let Some(version) = object.get("schema_version") {
let version = version.as_u64().ok_or_else(|| {
serde_json::Error::io(std::io::Error::new(
std::io::ErrorKind::InvalidData,
"resolved Worker manifest snapshot schema_version must be an integer",
))
})?;
if version != RESOLVED_MANIFEST_SNAPSHOT_SCHEMA_VERSION {
return Err(serde_json::Error::io(std::io::Error::new(
std::io::ErrorKind::InvalidData,
format!("unsupported resolved Worker manifest snapshot schema version {version}"),
)));
}
if object.len() != 2 {
return Err(serde_json::Error::io(std::io::Error::new(
std::io::ErrorKind::InvalidData,
"resolved Worker manifest snapshot contains unknown fields",
)));
}
let manifest = object.get("manifest").cloned().ok_or_else(|| {
serde_json::Error::io(std::io::Error::new(
std::io::ErrorKind::InvalidData,
"resolved Worker manifest snapshot is missing manifest",
))
})?;
if manifest
.as_object()
.is_some_and(|manifest| manifest.contains_key("memory"))
{
return Err(serde_json::Error::io(std::io::Error::new(
std::io::ErrorKind::InvalidData,
"current resolved Worker manifest contains removed top-level memory authority",
)));
}
return validate_persisted_worker_manifest(serde_json::from_value(manifest)?);
}
migrate_legacy_resolved_manifest_snapshot(snapshot)
}
fn validate_persisted_worker_manifest(
manifest: WorkerManifest,
) -> Result<WorkerManifest, serde_json::Error> {
manifest
.feature
.memory
.validate_execution()
.map_err(|message| {
serde_json::Error::io(std::io::Error::new(
std::io::ErrorKind::InvalidData,
message,
))
})?;
Ok(manifest)
}
fn migrate_legacy_resolved_manifest_snapshot(
mut snapshot: serde_json::Value,
) -> Result<WorkerManifest, serde_json::Error> {
let root = snapshot.as_object_mut().ok_or_else(|| {
serde_json::Error::io(std::io::Error::new(
std::io::ErrorKind::InvalidData,
"legacy resolved Worker manifest snapshot must be an object",
))
})?;
let legacy_memory = root.remove("memory");
let feature = root
.entry("feature")
.or_insert_with(|| serde_json::json!({}))
.as_object_mut()
.ok_or_else(|| {
serde_json::Error::io(std::io::Error::new(
std::io::ErrorKind::InvalidData,
"legacy resolved Worker manifest feature must be an object",
))
})?;
let legacy_feature_memory: LegacyMemoryFeatureConfig = serde_json::from_value(
feature
.remove("memory")
.unwrap_or_else(|| serde_json::json!({})),
)?;
let enabled = legacy_feature_memory.enabled;
let staging_tools = legacy_feature_memory.staging;
let legacy_memory: LegacyMemoryConfig =
serde_json::from_value(legacy_memory.unwrap_or_else(|| serde_json::json!({})))?;
let mut workspace_settings = match (
legacy_memory.workspace_id,
legacy_memory.settings_revision,
legacy_memory.language,
) {
(Some(workspace_id), Some(settings_revision), Some(language)) => Some(serde_json::json!({
"workspace_id": workspace_id,
"settings_revision": settings_revision,
"language": language,
})),
(None, None, None) => None,
_ => {
return Err(serde_json::Error::io(std::io::Error::new(
std::io::ErrorKind::InvalidData,
"legacy resolved Worker manifest contains a partial Memory settings snapshot",
)));
}
};
if !enabled {
workspace_settings = None;
}
let extraction_enabled = legacy_memory.extract_threshold.is_some();
if legacy_memory.consolidation_model.is_some() {
return Err(serde_json::Error::io(std::io::Error::new(
std::io::ErrorKind::InvalidData,
"legacy resolved Worker manifest uses a Worker-owned consolidation model that cannot be migrated to Backend authority",
)));
}
let consolidation_enabled = match (
legacy_memory.consolidation_threshold_files,
legacy_memory.consolidation_threshold_bytes,
) {
(None, None) => false,
(Some(5), Some(50_000)) => true,
_ => {
return Err(serde_json::Error::io(std::io::Error::new(
std::io::ErrorKind::InvalidData,
"legacy resolved Worker manifest uses custom consolidation thresholds that cannot be migrated to Backend policy",
)));
}
};
let mut resolved = serde_json::json!({
"profile": {
"enabled": enabled,
"staging_tools": staging_tools,
"resident": {
"inject_summary": legacy_memory.inject_summary.unwrap_or(true),
},
"extraction": {
"enabled": extraction_enabled,
"model": serde_json::to_value(legacy_memory.extract_model)?,
"threshold": legacy_memory.extract_threshold,
"worker_max_turns": legacy_memory.extract_worker_max_turns,
},
"consolidation": {
"request_enabled": consolidation_enabled,
},
},
});
if let Some(workspace_settings) = workspace_settings {
resolved
.as_object_mut()
.expect("resolved Memory config is an object")
.insert("workspace_settings".to_string(), workspace_settings);
}
feature.insert("memory".to_string(), resolved);
validate_persisted_worker_manifest(serde_json::from_value(snapshot)?)
}
#[cfg(test)]
mod tests {
use super::*;
@@ -1246,36 +1467,182 @@ model_id = "claude-sonnet-4-20250514"
}
#[test]
fn omitted_memory_is_none() {
fn omitted_memory_feature_is_disabled() {
let manifest = WorkerManifest::from_toml(MINIMAL_REQUIRED).unwrap();
assert!(manifest.memory.is_none());
assert!(!manifest.feature.memory.profile.enabled);
assert!(manifest.feature.memory.workspace_settings.is_none());
}
#[test]
fn empty_memory_section_enables_with_default_root() {
let toml = format!("{MINIMAL_REQUIRED}\n[memory]\n");
fn resolved_memory_feature_requires_nested_profile_and_trusted_snapshot() {
let toml = format!(
"{MINIMAL_REQUIRED}\n\
[feature.memory.profile]\n\
enabled = true\n\
staging_tools = false\n\n\
[feature.memory.profile.resident]\n\
inject_summary = false\n\n\
[feature.memory.profile.extraction]\n\
enabled = true\n\
threshold = 42000\n\
worker_max_turns = 2\n\n\
[feature.memory.workspace_settings]\n\
workspace_id = \"workspace-1\"\n\
settings_revision = 7\n\
language = \"日本語\"\n"
);
let manifest = WorkerManifest::from_toml(&toml).unwrap();
let mem = manifest.memory.expect("memory section parsed");
assert!(mem.workspace_root.is_none());
assert_eq!(mem.inject_summary, None);
}
#[test]
fn memory_section_with_inject_summary_false() {
let toml = format!("{MINIMAL_REQUIRED}\n[memory]\ninject_summary = false\n");
let manifest = WorkerManifest::from_toml(&toml).unwrap();
let mem = manifest.memory.unwrap();
assert_eq!(mem.inject_summary, Some(false));
}
#[test]
fn memory_section_with_explicit_root() {
let toml = format!("{MINIMAL_REQUIRED}\n[memory]\nworkspace_root = \"/some/where\"\n");
let manifest = WorkerManifest::from_toml(&toml).unwrap();
let mem = manifest.memory.unwrap();
assert!(manifest.feature.memory.profile.enabled);
assert!(!manifest.feature.memory.profile.resident.inject_summary);
assert_eq!(
mem.workspace_root.unwrap(),
std::path::PathBuf::from("/some/where")
manifest.feature.memory.profile.extraction.threshold,
Some(42_000)
);
assert_eq!(
manifest
.feature
.memory
.workspace_settings()
.unwrap()
.language,
"日本語"
);
}
#[test]
fn resolved_memory_execution_validation_fails_closed() {
let snapshot = WorkspaceMemorySettingsSnapshot {
workspace_id: "workspace-1".to_string(),
settings_revision: 1,
language: "English".to_string(),
};
let mut enabled = ResolvedMemoryFeatureConfig::default();
enabled.profile.enabled = true;
assert!(enabled.validate_execution().is_err());
enabled.bind_workspace_settings(snapshot.clone()).unwrap();
assert!(enabled.validate_execution().is_ok());
let mut disabled = ResolvedMemoryFeatureConfig::default();
disabled.workspace_settings = Some(snapshot.clone());
assert!(disabled.validate_execution().is_err());
assert!(disabled.bind_workspace_settings(snapshot).is_err());
}
#[test]
fn current_manifest_rejects_legacy_top_level_memory_authority() {
let toml = format!("{MINIMAL_REQUIRED}\n[memory]\nlanguage = \"Japanese\"\n");
assert!(WorkerManifest::from_toml(&toml).is_err());
}
#[test]
fn persisted_manifest_adapter_migrates_legacy_memory_authority() {
let mut manifest =
serde_json::to_value(WorkerManifest::from_toml(MINIMAL_REQUIRED).unwrap()).unwrap();
manifest["feature"]["memory"] = serde_json::json!({
"enabled": true,
"staging": true,
});
manifest["memory"] = serde_json::json!({
"workspace_root": "/discarded",
"query_result_limit": 999,
"inject_summary": false,
"workspace_id": "workspace-1",
"settings_revision": 9,
"language": "Français",
"extract_threshold": 1234,
"extract_worker_max_turns": 3,
"consolidation_threshold_files": 5,
"consolidation_threshold_bytes": 50000,
});
let migrated = read_persisted_worker_manifest_snapshot(manifest).unwrap();
assert!(migrated.feature.memory.profile.enabled);
assert!(migrated.feature.memory.profile.staging_tools);
assert!(!migrated.feature.memory.profile.resident.inject_summary);
assert_eq!(
migrated.feature.memory.profile.extraction.threshold,
Some(1234)
);
assert!(
migrated
.feature
.memory
.profile
.consolidation
.request_enabled
);
assert_eq!(
migrated
.feature
.memory
.workspace_settings()
.unwrap()
.language,
"Français"
);
let current = write_persisted_worker_manifest_snapshot(&migrated).unwrap();
assert_eq!(current["schema_version"], 2);
assert!(current["manifest"].get("memory").is_none());
let mut disabled =
serde_json::to_value(WorkerManifest::from_toml(MINIMAL_REQUIRED).unwrap()).unwrap();
disabled["feature"]["memory"] = serde_json::json!({ "enabled": false });
disabled["memory"] = serde_json::json!({
"workspace_id": "workspace-1",
"settings_revision": 9,
"language": "Français",
});
let disabled = read_persisted_worker_manifest_snapshot(disabled).unwrap();
assert!(!disabled.feature.memory.profile.enabled);
assert!(disabled.feature.memory.workspace_settings.is_none());
}
#[test]
fn persisted_manifest_adapter_rejects_mixed_or_future_authority() {
let manifest =
serde_json::to_value(WorkerManifest::from_toml(MINIMAL_REQUIRED).unwrap()).unwrap();
let mut mixed = manifest.clone();
mixed["feature"]["memory"] = serde_json::json!({ "enabled": true, "profile": {} });
mixed["memory"] = serde_json::json!({});
assert!(read_persisted_worker_manifest_snapshot(mixed).is_err());
let mut custom_policy =
serde_json::to_value(WorkerManifest::from_toml(MINIMAL_REQUIRED).unwrap()).unwrap();
custom_policy["feature"]["memory"] = serde_json::json!({ "enabled": true });
custom_policy["memory"] = serde_json::json!({
"workspace_id": "workspace-1",
"settings_revision": 1,
"language": "English",
"consolidation_threshold_files": 99,
"consolidation_threshold_bytes": 50000,
});
assert!(read_persisted_worker_manifest_snapshot(custom_policy).is_err());
let current = WorkerManifest::from_toml(MINIMAL_REQUIRED).unwrap();
let mut current = write_persisted_worker_manifest_snapshot(&current).unwrap();
current["manifest"]["memory"] = serde_json::json!({
"workspace_id": "workspace-1",
"settings_revision": 1,
"language": "English",
});
assert!(read_persisted_worker_manifest_snapshot(current).is_err());
let mut missing_settings = WorkerManifest::from_toml(MINIMAL_REQUIRED).unwrap();
missing_settings.feature.memory.profile.enabled = true;
let missing_settings = write_persisted_worker_manifest_snapshot(&missing_settings).unwrap();
assert!(read_persisted_worker_manifest_snapshot(missing_settings).is_err());
let mut malformed_legacy = manifest.clone();
malformed_legacy["feature"]["memory"] = serde_json::json!({ "enabled": "yes" });
malformed_legacy["memory"] = serde_json::json!({ "unknown": true });
assert!(read_persisted_worker_manifest_snapshot(malformed_legacy).is_err());
assert!(
read_persisted_worker_manifest_snapshot(serde_json::json!({
"schema_version": 3,
"manifest": manifest,
}))
.is_err()
);
}
@@ -1291,14 +1658,6 @@ model_id = "claude-sonnet-4-20250514"
));
}
#[test]
fn memory_section_with_language() {
let toml = format!("{MINIMAL_REQUIRED}\n[memory]\nlanguage = \"Japanese\"\n");
let manifest = WorkerManifest::from_toml(&toml).unwrap();
let mem = manifest.memory.unwrap();
assert_eq!(mem.language.as_deref(), Some("Japanese"));
}
#[test]
fn reject_unknown_scheme() {
let toml =
+9 -61
View File
@@ -20,9 +20,9 @@ use crate::config::{
use crate::model::{AuthRef, ModelManifest};
use crate::plugin::PluginConfig;
use crate::{
EngineManifestConfig, McpConfig, McpStdioCwdPolicy, MemoryConfig, Permission, ResolveError,
ScopeConfig, ScopeRule, SkillsConfig, WebConfig, WorkerManifest, WorkerManifestConfig,
WorkerMetaConfig, paths,
EngineManifestConfig, McpConfig, McpStdioCwdPolicy, Permission, ResolveError, ScopeConfig,
ScopeRule, SkillsConfig, WebConfig, WorkerManifest, WorkerManifestConfig, WorkerMetaConfig,
paths,
};
const PROFILE_FORMAT_V1: &str = "yoi.profile.v1";
@@ -185,7 +185,7 @@ pub fn validate_profile_execution_target(
if feature.manage_workdir.enabled {
requirements.insert(WorkspaceAuthorityRequirement::ManageWorkdir);
}
if feature.memory.enabled || feature.memory.staging {
if feature.memory.profile.enabled || feature.memory.profile.staging_tools {
requirements.insert(WorkspaceAuthorityRequirement::Memory);
}
if feature.merge_request.show
@@ -642,7 +642,6 @@ fn resolve_profile_value(
mcp: profile.mcp,
compaction,
web: profile.web,
memory: profile.memory.map(Into::into),
skills: profile.skills,
};
let config =
@@ -663,51 +662,6 @@ fn resolve_profile_value(
})
}
#[derive(Debug, Default, Deserialize)]
#[serde(deny_unknown_fields)]
struct ProfileMemoryConfig {
#[serde(default)]
workspace_root: Option<PathBuf>,
#[serde(default)]
query_result_limit: Option<usize>,
#[serde(default)]
query_excerpt_lines: Option<usize>,
#[serde(default)]
inject_summary: Option<bool>,
#[serde(default)]
extract_model: Option<ModelManifest>,
#[serde(default)]
extract_threshold: Option<u64>,
#[serde(default)]
extract_worker_max_turns: Option<u32>,
#[serde(default)]
consolidation_model: Option<ModelManifest>,
#[serde(default)]
consolidation_threshold_files: Option<usize>,
#[serde(default)]
consolidation_threshold_bytes: Option<u64>,
}
impl From<ProfileMemoryConfig> for MemoryConfig {
fn from(profile: ProfileMemoryConfig) -> Self {
Self {
workspace_root: profile.workspace_root,
query_result_limit: profile.query_result_limit,
query_excerpt_lines: profile.query_excerpt_lines,
inject_summary: profile.inject_summary,
workspace_id: None,
settings_revision: None,
language: None,
extract_model: profile.extract_model,
extract_threshold: profile.extract_threshold,
extract_worker_max_turns: profile.extract_worker_max_turns,
consolidation_model: profile.consolidation_model,
consolidation_threshold_files: profile.consolidation_threshold_files,
consolidation_threshold_bytes: profile.consolidation_threshold_bytes,
}
}
}
#[derive(Debug, Default, Deserialize)]
#[serde(deny_unknown_fields)]
struct ProfileConfig {
@@ -738,8 +692,6 @@ struct ProfileConfig {
#[serde(default)]
web: Option<WebConfig>,
#[serde(default)]
memory: Option<ProfileMemoryConfig>,
#[serde(default)]
skills: Option<SkillsConfig>,
}
@@ -940,12 +892,6 @@ fn validate_profile_paths(profile: &ProfileConfig) -> Result<(), ProfileError> {
.map_err(|source| ProfileError::ProfileDeserialize { source })?;
reject_absolute_auth_file(&model.auth, "compaction.model.auth.file")?;
}
if let Some(memory) = &profile.memory
&& let Some(root) = &memory.workspace_root
&& root.is_absolute()
{
return Err(ProfileError::InvalidProfile("field `memory.workspace_root` is a resolved path and is not allowed in reusable Profiles".into()));
}
if let Some(skills) = &profile.skills {
for dir in &skills.directories {
if dir.is_absolute() {
@@ -1299,7 +1245,9 @@ mod tests {
("settings_revision", serde_json::json!(2)),
("language", serde_json::json!("Japanese")),
] {
let artifact = serde_json::json!({ "memory": { (field): value } });
let artifact = serde_json::json!({
"feature": { "memory": { (field): value } }
});
let error = resolve_profile_artifact_value(
artifact,
ProfileSource::Registry {
@@ -1351,7 +1299,7 @@ mod tests {
assert!(resolved.manifest.delegation_scope.allow.iter().any(|rule| {
rule.permission == protocol::Permission::Write && rule.target == tmp.path()
}));
assert!(!resolved.manifest.feature.memory.enabled);
assert!(!resolved.manifest.feature.memory.profile.enabled);
assert!(!resolved.manifest.feature.ticket.enabled);
assert!(!resolved.manifest.feature.objective.enabled);
assert!(!resolved.manifest.feature.flow.enabled);
@@ -1630,7 +1578,7 @@ enabled = false
.unwrap();
assert_eq!(resolved.manifest.worker.name, "runtime-worker");
assert!(resolved.manifest.feature.task.enabled);
assert!(!resolved.manifest.feature.memory.enabled);
assert!(!resolved.manifest.feature.memory.profile.enabled);
assert!(resolved.manifest.feature.web.enabled);
assert!(resolved.manifest.feature.sub_worker.enabled);
assert!(resolved.manifest.feature.ticket.enabled);
+13 -5
View File
@@ -152,13 +152,10 @@ pub enum MemoryStagingAffectedMemoryOperation {
}
#[derive(Debug, Clone, Serialize, Deserialize)]
#[serde(deny_unknown_fields)]
pub struct MemoryConsolidateStagingOperation {
#[serde(default)]
pub force: bool,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub threshold_files: Option<usize>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub threshold_bytes: Option<u64>,
}
#[derive(Debug, Clone, Serialize, Deserialize)]
@@ -450,10 +447,21 @@ mod tests {
use super::*;
use crate::extract::{CandidateKind, ExtractedCandidate};
#[test]
fn consolidation_operation_rejects_caller_owned_thresholds() {
let error =
serde_json::from_value::<MemoryConsolidateStagingOperation>(serde_json::json!({
"force": false,
"threshold_files": 1,
}))
.unwrap_err();
assert!(error.to_string().contains("threshold_files"));
}
#[test]
fn staging_list_read_close_records_reason_and_deletes_candidate() {
let temp = tempfile::tempdir().unwrap();
let layout = WorkspaceLayout::resolve(&manifest::MemoryConfig::default(), temp.path());
let layout = WorkspaceLayout::resolve(temp.path());
let source = SourceRef {
segment_id: "segment-1".into(),
range: [0, 1],
+1 -2
View File
@@ -21,8 +21,7 @@ pub struct StagingEntry {
pub id: Uuid,
pub path: PathBuf,
pub record: StagingRecord,
/// このファイルのバイト長。閾値判定 (`consolidation_threshold_bytes`)
/// に使う。
/// このファイルのバイト長。Backendのconsolidation閾値判定に使用する。
pub bytes: u64,
}
+8 -33
View File
@@ -70,24 +70,12 @@ impl WorkspaceLayout {
Self { root: root.into() }
}
/// Resolve a layout from a `MemoryConfig`.
/// Resolve a layout from the nearest Memory marker.
///
/// An explicit `memory.workspace_root` is honored exactly. Without an
/// explicit root, resolution searches `default_root` and its ancestors for
/// the nearest `.yoi/memory` directory. This keeps child worktrees that
/// contain `.yoi` project records such as tickets from
/// becoming independent memory roots merely because they contain `.yoi`.
///
/// If no memory marker exists, this falls back to `default_root` because
/// existing call sites require a concrete layout. That fallback is a
/// no-marker compatibility path, not a `.yoi` marker interpretation; it
/// must not be used as evidence that `.yoi` alone enables repo-local
/// memory.
pub fn resolve(cfg: &manifest::MemoryConfig, default_root: &Path) -> Self {
if let Some(root) = &cfg.workspace_root {
return Self::new(root.clone());
}
/// Resolution searches `default_root` and its ancestors for the nearest
/// `.yoi/memory` directory. This legacy local-storage helper owns its path
/// policy directly; resolved Worker Manifests do not carry storage paths.
pub fn resolve(default_root: &Path) -> Self {
let root =
find_memory_marker_root(default_root).unwrap_or_else(|| default_root.to_path_buf());
Self::new(root)
@@ -335,16 +323,6 @@ mod tests {
assert!(matches!(err, LintError::InvalidPath(_)));
}
#[test]
fn resolve_uses_workspace_root_when_set() {
let cfg = manifest::MemoryConfig {
workspace_root: Some(PathBuf::from("/explicit")),
..Default::default()
};
let layout = WorkspaceLayout::resolve(&cfg, Path::new("/fallback"));
assert_eq!(layout.root(), Path::new("/explicit"));
}
#[test]
fn resolve_selects_nearest_ancestor_memory_marker_when_workspace_root_missing() {
let tmp = TempDir::new().unwrap();
@@ -353,8 +331,7 @@ mod tests {
std::fs::create_dir_all(workspace.join(".yoi/memory")).unwrap();
std::fs::create_dir_all(&child).unwrap();
let cfg = manifest::MemoryConfig::default();
let layout = WorkspaceLayout::resolve(&cfg, &child);
let layout = WorkspaceLayout::resolve(&child);
assert_eq!(layout.root(), workspace.as_path());
}
@@ -366,8 +343,7 @@ mod tests {
std::fs::create_dir_all(workspace.join(".yoi/memory")).unwrap();
std::fs::create_dir_all(child.join(".yoi/tickets")).unwrap();
let cfg = manifest::MemoryConfig::default();
let layout = WorkspaceLayout::resolve(&cfg, &child);
let layout = WorkspaceLayout::resolve(&child);
assert_eq!(layout.root(), workspace.as_path());
}
@@ -381,8 +357,7 @@ mod tests {
assert_eq!(find_memory_marker_root(&child), None);
let cfg = manifest::MemoryConfig::default();
let layout = WorkspaceLayout::resolve(&cfg, &child);
let layout = WorkspaceLayout::resolve(&child);
assert_eq!(layout.root(), child.as_path());
}
}
+9 -5
View File
@@ -3,7 +3,7 @@ mod common;
use std::ops::{Deref, DerefMut};
use std::sync::Arc;
use agen::interceptor::{Interceptor, TurnEndAction};
use agen::interceptor::{AssistantTurnEndContext, Interceptor, InterceptorResult, TurnEndAction};
use agen::llm_client::event::{Event, ResponseStatus, StatusEvent};
use agen::llm_client::types::{Item, RequestConfig};
use agen::tool::{Tool, ToolDefinition, ToolError, ToolMeta, ToolOutput};
@@ -100,8 +100,11 @@ struct PausePolicy;
#[async_trait]
impl Interceptor for PausePolicy {
async fn on_turn_end(&self, _history: &[Item]) -> TurnEndAction {
TurnEndAction::Pause
async fn on_assistant_turn_end(
&self,
_context: AssistantTurnEndContext<'_>,
) -> InterceptorResult<TurnEndAction> {
Ok(TurnEndAction::Pause)
}
}
@@ -350,7 +353,8 @@ async fn session_run_with_tool_call() {
async fn session_resume_after_pause() {
let (_dir, store) = make_store();
// First run: tool call with pause policy → Paused
// First terminal assistant response requests a tool; the assistant-turn
// interceptor pauses before the Engine enters the tool phase.
let client = MockLlmClient::with_responses(tool_call_events());
let mut worker = TestWorker::new(Engine::new(client));
worker.register_tool(weather_tool_definition());
@@ -386,7 +390,7 @@ async fn session_resume_after_pause() {
// Restore state and verify
let state = session_store::restore(&store, sid, segid).unwrap();
assert!(state.last_run_interrupted);
assert_eq!(state.active_run_turn_count, Some(2));
assert_eq!(state.active_run_turn_count, Some(1));
}
#[tokio::test]
+1 -6
View File
@@ -228,7 +228,7 @@ worker_context_max_tokens = 100000
enabled = true
[feature.memory]
enabled = true
enabled = false
[feature.web]
enabled = true
@@ -241,11 +241,6 @@ enabled = true
authoring = true
thread = true
[memory]
extract_threshold = 50000
consolidation_threshold_files = 5
consolidation_threshold_bytes = 50000
[web]
enabled = true
+9 -15
View File
@@ -759,8 +759,8 @@ fn migrate_worker_aggregate_document(
.get_mut("resolved_manifest_snapshot")
.filter(|snapshot| !snapshot.is_null())
{
let manifest: manifest::WorkerManifest =
serde_json::from_value(snapshot.clone()).map_err(|error| {
let mut manifest = manifest::read_persisted_worker_manifest_snapshot(snapshot.clone())
.map_err(|error| {
runtime_store_corrupt(
metadata_path,
format!("decode Worker aggregate resolved manifest snapshot: {error}"),
@@ -775,20 +775,14 @@ fn migrate_worker_aggregate_document(
),
));
}
snapshot
.as_object_mut()
.and_then(|manifest| manifest.get_mut("worker"))
.and_then(serde_json::Value::as_object_mut)
.ok_or_else(|| {
manifest.worker.name = expected_name.clone();
*snapshot =
manifest::write_persisted_worker_manifest_snapshot(&manifest).map_err(|error| {
runtime_store_corrupt(
metadata_path,
"Worker aggregate resolved manifest is missing worker metadata".to_string(),
format!("encode migrated Worker aggregate resolved manifest: {error}"),
)
})?
.insert(
"name".to_string(),
serde_json::Value::String(expected_name.clone()),
);
})?;
}
metadata.insert(
"worker_name".to_string(),
@@ -809,8 +803,8 @@ fn migrate_worker_aggregate_document(
));
}
if let Some(snapshot) = metadata.resolved_manifest_snapshot {
let manifest: manifest::WorkerManifest =
serde_json::from_value(snapshot).map_err(|error| {
let manifest =
manifest::read_persisted_worker_manifest_snapshot(snapshot).map_err(|error| {
runtime_store_corrupt(
metadata_path,
format!("decode migrated Worker aggregate resolved manifest: {error}"),
+20 -6
View File
@@ -746,9 +746,15 @@ fn bind_workspace_memory_settings(
));
}
manifest
.feature
.memory
.get_or_insert_with(manifest::MemoryConfig::default)
.bind_workspace_settings(snapshot);
.bind_workspace_settings(snapshot.clone())
.map_err(str::to_string)?;
manifest
.feature
.memory
.validate_execution()
.map_err(str::to_string)?;
Ok(())
}
@@ -759,10 +765,18 @@ fn validate_worker_memory_settings(
let Some(expected) = request.memory_settings.as_ref() else {
return Ok(());
};
let actual = manifest
manifest
.feature
.memory
.as_ref()
.and_then(manifest::MemoryConfig::workspace_settings)
.validate_execution()
.map_err(str::to_string)?;
if !manifest.feature.memory.profile.enabled {
return Ok(());
}
let actual = manifest
.feature
.memory
.workspace_settings()
.ok_or_else(|| {
"Workspace Worker restored without its bound Memory settings snapshot".to_string()
})?;
@@ -3165,7 +3179,7 @@ mod tests {
Some(session_store::WorkerActiveSegmentRef::pending_segment(
session_id,
)),
Some(serde_json::to_value(&manifest).unwrap()),
Some(manifest::write_persisted_worker_manifest_snapshot(&manifest).unwrap()),
)
.unwrap();
+78 -19
View File
@@ -22,7 +22,10 @@ use std::sync::atomic::{AtomicBool, AtomicU64, Ordering};
use std::sync::{Arc, Mutex};
use agen::Item;
use agen::interceptor::{Interceptor, PreRequestAction, PreToolAction, ToolCallInfo};
use agen::interceptor::{
Interceptor, InterceptorResult, PreLlmRequestContext, PreRequestAction, PreToolAction,
ToolCallInfo,
};
use agen::tool::{Tool, ToolDefinition, ToolError, ToolMeta, ToolOutput, ToolResult};
use async_trait::async_trait;
use serde::Deserialize;
@@ -397,15 +400,19 @@ impl CompactWorkerInterceptor {
}
#[async_trait]
impl Interceptor for CompactWorkerInterceptor {
async fn pre_llm_request(&self, context: &mut Vec<Item>) -> PreRequestAction {
impl<A: Send + Sync> Interceptor<A> for CompactWorkerInterceptor {
async fn pre_llm_request(
&self,
context: PreLlmRequestContext<'_, A>,
) -> InterceptorResult<PreRequestAction> {
let context = context.items;
let records = self.usage_tracker.records();
let estimate = agen::token_counter::total_tokens(context, &records);
if estimate.tokens > self.max_input_tokens {
return PreRequestAction::Cancel(format!(
return Ok(PreRequestAction::Cancel(format!(
"compact worker input occupancy exceeded {} tokens",
self.max_input_tokens
));
)));
}
let remaining = self.max_input_tokens.saturating_sub(estimate.tokens);
@@ -413,25 +420,28 @@ impl Interceptor for CompactWorkerInterceptor {
.store(remaining, Ordering::Release);
if let Some(item) = self.maybe_emit_warning(remaining) {
self.usage_tracker.note_request(context.len() + 1);
return PreRequestAction::ContinueWith(vec![item]);
return Ok(PreRequestAction::ContinueWith(vec![item]));
}
self.usage_tracker.note_request(context.len());
PreRequestAction::Continue
Ok(PreRequestAction::Continue)
}
async fn pre_tool_call(&self, info: &mut ToolCallInfo) -> PreToolAction {
async fn pre_tool_call(
&self,
info: &mut ToolCallInfo<'_, A>,
) -> InterceptorResult<PreToolAction> {
if self.final_reserve_tokens == 0 || info.call.name == "write_summary" {
return PreToolAction::Continue;
return Ok(PreToolAction::Continue);
}
let remaining = self.last_remaining_tokens.load(Ordering::Acquire);
if remaining > self.final_reserve_tokens {
return PreToolAction::Continue;
return Ok(PreToolAction::Continue);
}
PreToolAction::SyntheticResult(ToolResult::error(
Ok(PreToolAction::SyntheticResult(ToolResult::error(
info.call.id.clone(),
"compact worker final reserve reached; do not perform more exploratory tool reads. Call `write_summary` now.",
))
)))
}
}
@@ -467,13 +477,27 @@ mod tests {
let mut context = vec![Item::user_message("hello")];
assert!(matches!(
interceptor.pre_llm_request(&mut context).await,
interceptor
.pre_llm_request(PreLlmRequestContext::<()> {
invocation: Default::default(),
items: &mut context,
history: &[],
})
.await
.unwrap(),
PreRequestAction::Continue
));
tracker.record_usage(&make_usage(100));
assert!(matches!(
interceptor.pre_llm_request(&mut context).await,
interceptor
.pre_llm_request(PreLlmRequestContext::<()> {
invocation: Default::default(),
items: &mut context,
history: &[],
})
.await
.unwrap(),
PreRequestAction::Continue
));
tracker.record_usage(&make_usage(100));
@@ -481,7 +505,14 @@ mod tests {
// Two 100-token requests would exceed a cumulative 150-token cap, but
// current occupancy is still the latest 100-token measurement.
assert!(matches!(
interceptor.pre_llm_request(&mut context).await,
interceptor
.pre_llm_request(PreLlmRequestContext::<()> {
invocation: Default::default(),
items: &mut context,
history: &[],
})
.await
.unwrap(),
PreRequestAction::Continue
));
}
@@ -503,13 +534,27 @@ mod tests {
let mut context = vec![Item::user_message("hello")];
assert!(matches!(
interceptor.pre_llm_request(&mut context).await,
interceptor
.pre_llm_request(PreLlmRequestContext::<()> {
invocation: Default::default(),
items: &mut context,
history: &[],
})
.await
.unwrap(),
PreRequestAction::Continue
));
tracker.record_usage(&make_usage(100));
assert!(matches!(
interceptor.pre_llm_request(&mut context).await,
interceptor
.pre_llm_request(PreLlmRequestContext::<()> {
invocation: Default::default(),
items: &mut context,
history: &[],
})
.await
.unwrap(),
PreRequestAction::ContinueWith(items)
if items.len() == 1 && items[0].as_text().unwrap_or_default().contains("write_summary")
));
@@ -523,13 +568,27 @@ mod tests {
let mut context = vec![Item::user_message("hello")];
assert!(matches!(
interceptor.pre_llm_request(&mut context).await,
interceptor
.pre_llm_request(PreLlmRequestContext::<()> {
invocation: Default::default(),
items: &mut context,
history: &[],
})
.await
.unwrap(),
PreRequestAction::Continue
));
tracker.record_usage(&make_usage(100));
assert!(matches!(
interceptor.pre_llm_request(&mut context).await,
interceptor
.pre_llm_request(PreLlmRequestContext::<()> {
invocation: Default::default(),
items: &mut context,
history: &[],
})
.await
.unwrap(),
PreRequestAction::Cancel(message) if message.contains("occupancy")
));
}
+48 -43
View File
@@ -197,8 +197,8 @@ async fn finish_controller_run<C, St>(
{
// history / user_segments are no longer mirrored on WorkerSharedState —
// clients reconstruct them from `Event::Snapshot` + live
// `Event::Entry` deliveries driven by the session-log sink. We
// flip the status and kick post-run memory jobs here.
// `Event::Entry` deliveries driven by the session-log sink. The
// lifecycle hook/task registry observes the terminal commit separately.
//
// In-flight blocks are run-local streaming state, not durable transcript.
// Any block not cleared by a committed AssistantItem must be discarded at
@@ -206,7 +206,6 @@ async fn finish_controller_run<C, St>(
// partial text/tool arguments after newer entries.
worker.clear_in_flight_events();
set_controller_status(shared_state, runtime_dir, working_event_tx, new_status).await;
worker.spawn_post_run_memory_jobs();
}
/// Pending turn launch staged by an event handler for the next outer-loop
@@ -938,7 +937,6 @@ where
let local_filesystem = worker.local_working_directory().cloned();
let local_workspace_root = local_filesystem.as_ref().map(|local| local.root.clone());
let task_feature = worker.task_feature();
let memory_config = worker.manifest().memory.clone();
let web_config = worker.manifest().web.clone();
let mcp_config = worker.manifest().mcp.clone();
let spawner_name = worker.manifest().worker.name.clone();
@@ -995,6 +993,41 @@ where
let worker_enabled = feature_config.worker.enabled;
let sub_worker_enabled = feature_config.sub_worker.enabled;
let mut feature_registry = FeatureRegistryBuilder::new();
let memory_install_plan = crate::feature::builtin::memory::MemoryFeatureInstallPlan::prepare(
worker.manifest(),
worker.workspace_client_handle(),
worker.prompts().load_full(),
)
.await?;
let memory_prompt_contribution = memory_install_plan.as_ref().map(|plan| {
(
plan.resident_summary.clone(),
plan.system_prompt_override.clone(),
)
});
let memory_lifecycle_config = memory_install_plan
.as_ref()
.map(|plan| plan.resolved_config.clone());
if let Some(plan) = memory_install_plan {
feature_registry.add_module(plan.module);
}
if let Some(memory_config) = memory_lifecycle_config
&& let Some(memory_lifecycle) =
crate::feature::builtin::memory_lifecycle::MemoryLifecycleFeature::from_resolved_config(
worker.manifest_lifecycle_features_enabled(),
memory_config,
worker.committed_session_capture_handle(),
worker.session_extension_handle(),
worker.workspace_client_handle(),
spawner_manifest.clone(),
worker.llm_client_handle(),
prompts.clone(),
spawner_workspace_context.clone(),
worker.working_event_sender(),
)?
{
feature_registry.add_module(memory_lifecycle);
}
if sub_worker_enabled && !worker_enabled {
feature_registry.add_module(
crate::feature::builtin::manage_worker::sub_worker_control_feature(
@@ -1148,38 +1181,6 @@ where
}
}
// Memory tools require explicit feature exposure. Workspace memory access
// is authority-bound to the Backend Workspace API; the Worker must not
// register local filesystem memory tools even when it has local cwd/root
// authority for shell/file tools.
if feature_config.memory.enabled {
let _mem = memory_config.as_ref().ok_or_else(|| {
std::io::Error::new(
std::io::ErrorKind::InvalidInput,
"[feature.memory].enabled = true requires a [memory] configuration section",
)
})?;
if workspace_client.is_available() && workspace_client.workspace_id().is_some() {
let definitions = if feature_config.memory.staging {
crate::feature::builtin::memory::workspace_http_memory_consolidation_tools(
workspace_client.clone(),
)
} else {
crate::feature::builtin::memory::workspace_http_memory_tools(
workspace_client.clone(),
)
};
for definition in definitions {
engine.register_tool(definition);
}
} else {
return Err(std::io::Error::new(
std::io::ErrorKind::InvalidInput,
"memory tools require Backend Workspace API authority",
));
}
}
let mut observation_providers: Vec<
Arc<dyn crate::feature::builtin::worker_observation::WorkerObservationProvider>,
> = Vec::new();
@@ -1235,6 +1236,9 @@ where
),
));
}
if let Some((resident_summary, system_prompt_override)) = memory_prompt_contribution {
worker.install_system_prompt_contribution(resident_summary, system_prompt_override);
}
if let Some(tracker) = tracker {
worker.attach_tracker(tracker);
}
@@ -1590,7 +1594,9 @@ async fn controller_loop<C, St>(
&working_event_tx,
target,
expected_head_entries,
) {
)
.await
{
worker.clear_in_flight_events();
shared_state.set_status(WorkerStatus::Idle);
let _ = working_event_tx.send(Event::Status {
@@ -1711,10 +1717,9 @@ async fn controller_loop<C, St>(
tracing::warn!(%error, "Worker runtime socket cleanup failed");
}
// Background memory jobs own extract/consolidate workers after a
// turn completes. Join them before closing the Workdir session so no
// Worker-owned task can outlive its operation attachment.
worker.wait_for_memory_jobs().await;
// Feature callbacks and tasks share the Worker scope. Stop them before
// Memory/Workdir teardown so they cannot observe a partially closed Worker.
worker.stop_feature_runtime("controller shutdown").await;
if let Some(session) = worker.workdir_session()
&& let Err(error) = session.close().await
@@ -1993,7 +1998,7 @@ where
}
}
fn apply_rewind<C, St>(
async fn apply_rewind<C, St>(
worker: &mut Worker<C, St>,
working_event_tx: &broadcast::Sender<Event>,
target: RewindTargetId,
@@ -2003,7 +2008,7 @@ where
C: LlmClient + 'static,
St: Store,
{
match worker.rewind_to(target, expected_head_entries) {
match worker.rewind_to(target, expected_head_entries).await {
Ok(applied) => {
let session =
session_store::public_snapshot::project_current_session_snapshot(&applied.entries);
+392 -148
View File
@@ -23,7 +23,14 @@ use agen::tool::ToolDefinition;
use serde::{Deserialize, Serialize};
use thiserror::Error;
use crate::hook::{Hook, HookRegistryBuilder, OnTurnEnd, PostToolCall, PreLlmRequest, PreToolCall};
use crate::hook::{
BeforeSessionRewrite, Hook, HookExecutionPolicy, HookRegistryBuilder, OnPromptSubmit,
OnTurnEnd, PostToolCall, PreLlmRequest, PreToolCall, RunCommitted, RunExit, WorkerStopping,
};
use background::{
BackgroundTaskSpec, FeatureBackgroundTask, FeatureBackgroundTaskRegistry,
FeatureBackgroundTaskRegistryBuilder,
};
/// Stable source-qualified identifier for a feature module.
#[derive(Clone, Debug, PartialEq, Eq, Hash, PartialOrd, Ord, Serialize, Deserialize)]
@@ -253,10 +260,15 @@ pub enum FeatureRuntimeKind {
#[derive(Clone, Debug, PartialEq, Eq, Hash, Serialize, Deserialize)]
#[serde(rename_all = "snake_case")]
pub enum FeatureHookPoint {
PreRequest,
PromptSubmit,
PreLlmRequest,
PreToolCall,
ToolResult,
TurnEnd,
PostToolCall,
AssistantTurnEnd,
RunExit,
RunCommitted,
BeforeSessionRewrite,
WorkerStopping,
}
/// Serializable declaration of a tool contribution. The executable factory is
@@ -379,16 +391,17 @@ impl FeatureInstructionContribution {
}
}
/// Background task lifecycle phase represented by this registry slice.
#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize)]
/// Background tasks are always Worker-managed and execute inside the owning
/// feature scope. Report-only and detached host-managed declarations are not
/// accepted by the current contract.
#[derive(Clone, Debug, PartialEq, Eq, PartialOrd, Ord, Serialize, Deserialize)]
#[serde(rename_all = "snake_case")]
pub enum BackgroundTaskLifecycle {
DescriptorOnly,
HostManaged,
WorkerManaged,
}
/// Declaration for a feature-provided background task.
#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize)]
/// Declaration for a feature-provided executable background task.
#[derive(Clone, Debug, PartialEq, Eq, PartialOrd, Ord, Serialize, Deserialize)]
pub struct BackgroundTaskDeclaration {
pub name: String,
pub description: String,
@@ -396,11 +409,11 @@ pub struct BackgroundTaskDeclaration {
}
impl BackgroundTaskDeclaration {
pub fn descriptor_only(name: impl Into<String>, description: impl Into<String>) -> Self {
pub fn worker_managed(name: impl Into<String>, description: impl Into<String>) -> Self {
Self {
name: name.into(),
description: description.into(),
lifecycle: BackgroundTaskLifecycle::DescriptorOnly,
lifecycle: BackgroundTaskLifecycle::WorkerManaged,
}
}
}
@@ -772,6 +785,15 @@ impl FeatureInstallReport {
}
}
fn clear_installed_contributions(&mut self) {
self.installed = false;
self.installed_tools.clear();
self.installed_hooks.clear();
self.installed_instructions.clear();
self.declared_background_tasks.clear();
self.provided_services.clear();
}
fn mark_skipped(
&mut self,
kind: FeatureContributionKind,
@@ -881,46 +903,6 @@ fn reject_undeclared_contribution(
error
}
/// Model-visible durable notification sink skeleton. The first slice exposes
/// the boundary without implementing a new event channel.
pub struct FeatureNotificationSink<'a> {
report: &'a mut FeatureInstallReport,
}
impl FeatureNotificationSink<'_> {
pub fn notify_model(&mut self, message: impl Into<String>) -> Result<(), FeatureInstallError> {
let message = message.into();
self.report.diagnostics.push(FeatureDiagnostic::warning(format!(
"model notification requested during feature installation but no durable Notify host is attached: {message}"
)));
self.report.mark_skipped(
FeatureContributionKind::Notification,
"notify_model",
"durable Notify/SystemItem host is not connected during feature installation",
);
Ok(())
}
}
/// Transient human-facing alert sink skeleton.
pub struct FeatureAlertSink<'a> {
report: &'a mut FeatureInstallReport,
}
impl FeatureAlertSink<'_> {
pub fn alert(&mut self, message: impl Into<String>) {
let message = message.into();
self.report
.diagnostics
.push(FeatureDiagnostic::info(format!("feature alert: {message}")));
self.report.mark_skipped(
FeatureContributionKind::Alert,
"alert",
"transient alert host is not connected during feature installation",
);
}
}
/// Diagnostic sink available to feature installers.
pub struct FeatureDiagnosticSink<'a> {
report: &'a mut FeatureInstallReport,
@@ -1042,15 +1024,74 @@ impl HookContributionRegistrar<'_> {
))
}
fn record(&mut self, declaration: HookDeclaration) {
if !self.report.installed_hooks.contains(&declaration) {
self.report.installed_hooks.push(declaration);
}
}
pub fn add_prompt_submit(
&mut self,
name: impl Into<String>,
policy: HookExecutionPolicy,
hook: impl Hook<OnPromptSubmit> + 'static,
) -> Result<(), FeatureInstallError> {
let declaration = HookDeclaration::new(name, FeatureHookPoint::PromptSubmit);
self.require_declared(&declaration)?;
self.hook_builder
.add_named_on_prompt_submit(
format!("{}:{}", self.feature_id, declaration.name),
policy,
hook,
)
.map_err(|error| FeatureInstallError::InvalidDescriptor(error.to_string()))?;
self.record(declaration);
Ok(())
}
pub fn add_pre_llm_request(
&mut self,
name: impl Into<String>,
policy: HookExecutionPolicy,
hook: impl Hook<PreLlmRequest> + 'static,
) -> Result<(), FeatureInstallError> {
let declaration = HookDeclaration::new(name, FeatureHookPoint::PreLlmRequest);
self.require_declared(&declaration)?;
self.hook_builder
.add_named_pre_llm_request(
format!("{}:{}", self.feature_id, declaration.name),
policy,
hook,
)
.map_err(|error| FeatureInstallError::InvalidDescriptor(error.to_string()))?;
self.record(declaration);
Ok(())
}
pub fn add_pre_request(
&mut self,
name: impl Into<String>,
hook: impl Hook<PreLlmRequest> + 'static,
) -> Result<(), FeatureInstallError> {
let declaration = HookDeclaration::new(name, FeatureHookPoint::PreRequest);
self.add_pre_llm_request(name, HookExecutionPolicy::fail_closed(), hook)
}
pub fn add_pre_tool_call_with_policy(
&mut self,
name: impl Into<String>,
policy: HookExecutionPolicy,
hook: impl Hook<PreToolCall> + 'static,
) -> Result<(), FeatureInstallError> {
let declaration = HookDeclaration::new(name, FeatureHookPoint::PreToolCall);
self.require_declared(&declaration)?;
self.hook_builder.add_pre_llm_request(hook);
self.report.installed_hooks.push(declaration);
self.hook_builder
.add_named_pre_tool_call(
format!("{}:{}", self.feature_id, declaration.name),
policy,
hook,
)
.map_err(|error| FeatureInstallError::InvalidDescriptor(error.to_string()))?;
self.record(declaration);
Ok(())
}
@@ -1059,10 +1100,25 @@ impl HookContributionRegistrar<'_> {
name: impl Into<String>,
hook: impl Hook<PreToolCall> + 'static,
) -> Result<(), FeatureInstallError> {
let declaration = HookDeclaration::new(name, FeatureHookPoint::PreToolCall);
self.add_pre_tool_call_with_policy(name, HookExecutionPolicy::fail_closed(), hook)
}
pub fn add_post_tool_call(
&mut self,
name: impl Into<String>,
policy: HookExecutionPolicy,
hook: impl Hook<PostToolCall> + 'static,
) -> Result<(), FeatureInstallError> {
let declaration = HookDeclaration::new(name, FeatureHookPoint::PostToolCall);
self.require_declared(&declaration)?;
self.hook_builder.add_pre_tool_call(hook);
self.report.installed_hooks.push(declaration);
self.hook_builder
.add_named_post_tool_call(
format!("{}:{}", self.feature_id, declaration.name),
policy,
hook,
)
.map_err(|error| FeatureInstallError::InvalidDescriptor(error.to_string()))?;
self.record(declaration);
Ok(())
}
@@ -1071,10 +1127,25 @@ impl HookContributionRegistrar<'_> {
name: impl Into<String>,
hook: impl Hook<PostToolCall> + 'static,
) -> Result<(), FeatureInstallError> {
let declaration = HookDeclaration::new(name, FeatureHookPoint::ToolResult);
self.add_post_tool_call(name, HookExecutionPolicy::fail_closed(), hook)
}
pub fn add_assistant_turn_end(
&mut self,
name: impl Into<String>,
policy: HookExecutionPolicy,
hook: impl Hook<OnTurnEnd> + 'static,
) -> Result<(), FeatureInstallError> {
let declaration = HookDeclaration::new(name, FeatureHookPoint::AssistantTurnEnd);
self.require_declared(&declaration)?;
self.hook_builder.add_post_tool_call(hook);
self.report.installed_hooks.push(declaration);
self.hook_builder
.add_named_on_turn_end(
format!("{}:{}", self.feature_id, declaration.name),
policy,
hook,
)
.map_err(|error| FeatureInstallError::InvalidDescriptor(error.to_string()))?;
self.record(declaration);
Ok(())
}
@@ -1083,10 +1154,82 @@ impl HookContributionRegistrar<'_> {
name: impl Into<String>,
hook: impl Hook<OnTurnEnd> + 'static,
) -> Result<(), FeatureInstallError> {
let declaration = HookDeclaration::new(name, FeatureHookPoint::TurnEnd);
self.add_assistant_turn_end(name, HookExecutionPolicy::fail_closed(), hook)
}
pub fn add_run_exit(
&mut self,
name: impl Into<String>,
policy: HookExecutionPolicy,
hook: impl Hook<RunExit> + 'static,
) -> Result<(), FeatureInstallError> {
let declaration = HookDeclaration::new(name, FeatureHookPoint::RunExit);
self.require_declared(&declaration)?;
self.hook_builder.add_on_turn_end(hook);
self.report.installed_hooks.push(declaration);
self.hook_builder
.add_named_run_exit(
format!("{}:{}", self.feature_id, declaration.name),
policy,
hook,
)
.map_err(|error| FeatureInstallError::InvalidDescriptor(error.to_string()))?;
self.record(declaration);
Ok(())
}
pub fn add_run_committed(
&mut self,
name: impl Into<String>,
policy: HookExecutionPolicy,
hook: impl Hook<RunCommitted> + 'static,
) -> Result<(), FeatureInstallError> {
let declaration = HookDeclaration::new(name, FeatureHookPoint::RunCommitted);
self.require_declared(&declaration)?;
self.hook_builder
.add_named_run_committed(
format!("{}:{}", self.feature_id, declaration.name),
policy,
hook,
)
.map_err(|error| FeatureInstallError::InvalidDescriptor(error.to_string()))?;
self.record(declaration);
Ok(())
}
pub fn add_before_session_rewrite(
&mut self,
name: impl Into<String>,
policy: HookExecutionPolicy,
hook: impl Hook<BeforeSessionRewrite> + 'static,
) -> Result<(), FeatureInstallError> {
let declaration = HookDeclaration::new(name, FeatureHookPoint::BeforeSessionRewrite);
self.require_declared(&declaration)?;
self.hook_builder
.add_named_before_session_rewrite(
format!("{}:{}", self.feature_id, declaration.name),
policy,
hook,
)
.map_err(|error| FeatureInstallError::InvalidDescriptor(error.to_string()))?;
self.record(declaration);
Ok(())
}
pub fn add_worker_stopping(
&mut self,
name: impl Into<String>,
policy: HookExecutionPolicy,
hook: impl Hook<WorkerStopping> + 'static,
) -> Result<(), FeatureInstallError> {
let declaration = HookDeclaration::new(name, FeatureHookPoint::WorkerStopping);
self.require_declared(&declaration)?;
self.hook_builder
.add_named_worker_stopping(
format!("{}:{}", self.feature_id, declaration.name),
policy,
hook,
)
.map_err(|error| FeatureInstallError::InvalidDescriptor(error.to_string()))?;
self.record(declaration);
Ok(())
}
}
@@ -1124,33 +1267,40 @@ impl FeatureInstructionRegistrar<'_> {
}
}
/// Background task registrar for descriptor/report-only contributions.
/// Registrar for executable, Worker-managed background task contributions.
pub struct BackgroundTaskRegistrar<'a> {
feature_id: &'a FeatureId,
declarations: &'a FeatureContributionDeclarations,
registry: &'a mut FeatureBackgroundTaskRegistryBuilder,
report: &'a mut FeatureInstallReport,
}
impl BackgroundTaskRegistrar<'_> {
pub fn declare(
pub fn register(
&mut self,
declaration: BackgroundTaskDeclaration,
spec: BackgroundTaskSpec,
task: impl FeatureBackgroundTask + 'static,
) -> Result<(), FeatureInstallError> {
if !self.declarations.contains_background_task(&declaration) {
if !self
.declarations
.contains_background_task(&spec.declaration)
{
return Err(reject_undeclared_contribution(
self.feature_id,
self.report,
FeatureContributionKind::BackgroundTask,
declaration.name,
spec.declaration.name,
));
}
self.registry
.register(self.feature_id.clone(), spec.clone(), task)?;
if !self
.report
.declared_background_tasks
.iter()
.any(|task| task.name == declaration.name)
.any(|task| task.name == spec.declaration.name)
{
self.report.declared_background_tasks.push(declaration);
self.report.declared_background_tasks.push(spec.declaration);
}
Ok(())
}
@@ -1330,15 +1480,17 @@ impl ProtocolProviderRegistrar<'_> {
}
}
for task in background_tasks {
if !self
.report
.declared_background_tasks
.iter()
.any(|declared| declared.name == task.name)
{
self.report.declared_background_tasks.push(task);
}
if let Some(task) = background_tasks.first() {
let reason = format!(
"protocol provider background task `{}` has no executable Worker-managed handler",
task.name
);
self.report.mark_skipped(
FeatureContributionKind::BackgroundTask,
task.name.clone(),
reason.clone(),
);
return Err(FeatureInstallError::InvalidDescriptor(reason));
}
Ok(())
@@ -1352,6 +1504,7 @@ pub struct FeatureInstallContext<'a> {
pending_tools: &'a mut Vec<ToolDefinition>,
installed_tool_names: &'a mut HashMap<String, FeatureId>,
hook_builder: &'a mut HookRegistryBuilder,
background_task_builder: &'a mut FeatureBackgroundTaskRegistryBuilder,
service_registry: &'a mut FeatureServiceRegistry,
report: &'a mut FeatureInstallReport,
}
@@ -1392,6 +1545,7 @@ impl FeatureInstallContext<'_> {
BackgroundTaskRegistrar {
feature_id: self.feature_id,
declarations: self.declarations,
registry: self.background_task_builder,
report: self.report,
}
}
@@ -1416,18 +1570,6 @@ impl FeatureInstallContext<'_> {
}
}
pub fn notifications(&mut self) -> FeatureNotificationSink<'_> {
FeatureNotificationSink {
report: self.report,
}
}
pub fn alerts(&mut self) -> FeatureAlertSink<'_> {
FeatureAlertSink {
report: self.report,
}
}
pub fn diagnostics(&mut self) -> FeatureDiagnosticSink<'_> {
FeatureDiagnosticSink {
report: self.report,
@@ -1440,6 +1582,7 @@ impl FeatureInstallContext<'_> {
pub struct FeatureRegistryInstallReport {
pub reports: Vec<FeatureInstallReport>,
pub services: FeatureServiceRegistry,
pub background_tasks: FeatureBackgroundTaskRegistry,
pub plan_error: Option<FeaturePlanError>,
}
@@ -1795,7 +1938,7 @@ impl FeatureRegistryBuilder {
}
/// Install modules into the existing Engine tool path and hook builder.
pub(crate) fn install_into_engine<C: LlmClient, A>(
pub(crate) fn install_into_engine<C: LlmClient, A: Send + Sync>(
self,
worker: &mut Engine<C, Mutable, A>,
hook_builder: &mut HookRegistryBuilder,
@@ -1861,12 +2004,16 @@ impl FeatureRegistryBuilder {
return FeatureRegistryInstallReport {
reports,
services: FeatureServiceRegistry::default(),
background_tasks: FeatureBackgroundTaskRegistry::default(),
plan_error: Some(error),
};
}
};
let mut service_registry = FeatureServiceRegistry::default();
let mut background_task_builder = FeatureBackgroundTaskRegistryBuilder::default();
let mut reports = Vec::with_capacity(plan.ordered_indices.len());
let install_hook_checkpoint = hook_builder.checkpoint();
let install_tool_checkpoint = pending_tools.len();
let mut modules = self.modules.into_iter().map(Some).collect::<Vec<_>>();
let ordered_modules = plan
.ordered_indices
@@ -1884,6 +2031,11 @@ impl FeatureRegistryBuilder {
for (module, descriptor) in ordered_modules {
let declarations = FeatureContributionDeclarations::from_descriptor(&descriptor);
let mut report = FeatureInstallReport::new(&descriptor);
let hook_checkpoint = hook_builder.checkpoint();
let background_checkpoint = background_task_builder.checkpoint();
let service_checkpoint = service_registry.clone();
let tool_checkpoint = pending_tools.len();
let installed_tool_checkpoint = installed_tool_names.clone();
let mut required_service_failed = false;
for requirement in descriptor.requires_services.iter().cloned() {
@@ -1920,10 +2072,6 @@ impl FeatureRegistryBuilder {
continue;
}
for background_task in descriptor.background_tasks.iter().cloned() {
report.declared_background_tasks.push(background_task);
}
let install_result = {
let mut context = FeatureInstallContext {
feature_id: &descriptor.id,
@@ -1931,6 +2079,7 @@ impl FeatureRegistryBuilder {
pending_tools,
installed_tool_names: &mut installed_tool_names,
hook_builder,
background_task_builder: &mut background_task_builder,
service_registry: &mut service_registry,
report: &mut report,
};
@@ -1940,18 +2089,81 @@ impl FeatureRegistryBuilder {
match install_result {
Ok(()) => report.installed = true,
Err(error) => {
hook_builder.rollback_to(hook_checkpoint);
background_task_builder.rollback_to(&background_checkpoint);
service_registry = service_checkpoint.clone();
pending_tools.truncate(tool_checkpoint);
installed_tool_names = installed_tool_checkpoint.clone();
report.clear_installed_contributions();
report
.diagnostics
.push(FeatureDiagnostic::error(error.to_string()));
}
}
if report.installed {
for hook in &descriptor.hooks {
if !report.installed_hooks.contains(hook) {
report.diagnostics.push(FeatureDiagnostic::error(format!(
"feature `{}` declared hook `{}` at {:?} but did not register it",
descriptor.id, hook.name, hook.point
)));
}
}
for task in &descriptor.background_tasks {
if !report.declared_background_tasks.contains(task) {
report.diagnostics.push(FeatureDiagnostic::error(format!(
"feature `{}` declared background task `{}` but did not register an executable handler",
descriptor.id, task.name
)));
}
}
if report
.diagnostics
.iter()
.any(|diagnostic| diagnostic.severity == FeatureDiagnosticSeverity::Error)
{
hook_builder.rollback_to(hook_checkpoint);
background_task_builder.rollback_to(&background_checkpoint);
service_registry = service_checkpoint.clone();
pending_tools.truncate(tool_checkpoint);
installed_tool_names = installed_tool_checkpoint.clone();
report.clear_installed_contributions();
report.clear_installed_contributions();
}
}
reports.push(report);
}
FeatureRegistryInstallReport {
reports,
services: service_registry,
plan_error: None,
let failed = reports.iter().any(|report| {
report
.diagnostics
.iter()
.any(|diagnostic| diagnostic.severity == FeatureDiagnosticSeverity::Error)
});
if failed {
hook_builder.rollback_to(install_hook_checkpoint);
pending_tools.truncate(install_tool_checkpoint);
for report in &mut reports {
if report.installed {
report.clear_installed_contributions();
report.diagnostics.push(FeatureDiagnostic::warning(
"feature scope rolled back because another contribution failed",
));
}
}
FeatureRegistryInstallReport {
reports,
services: FeatureServiceRegistry::default(),
background_tasks: FeatureBackgroundTaskRegistry::default(),
plan_error: None,
}
} else {
FeatureRegistryInstallReport {
reports,
services: service_registry,
background_tasks: background_task_builder.build(),
plan_error: None,
}
}
}
}
@@ -1996,9 +2208,11 @@ pub enum FeatureInstallError {
Install(String),
}
pub mod background;
pub mod builtin;
pub mod mcp;
pub mod plugin;
pub(crate) mod session;
#[cfg(test)]
mod tests {
@@ -2398,13 +2612,9 @@ mod tests {
}
#[test]
fn descriptor_contributions_are_recorded() {
fn executable_contributions_are_recorded() {
let descriptor = FeatureDescriptor::builtin("dummy", "Dummy")
.with_tool(ToolDeclaration::new("Dummy", "dummy tool"))
.with_background_task(BackgroundTaskDeclaration::descriptor_only(
"daily",
"descriptor-only background task",
));
.with_tool(ToolDeclaration::new("Dummy", "dummy tool"));
let mut hook_builder = HookRegistryBuilder::default();
let mut pending_tools = Vec::new();
let report = FeatureRegistryBuilder::new()
@@ -2420,7 +2630,7 @@ mod tests {
let feature_report = &report.reports[0];
assert!(feature_report.installed);
assert_eq!(feature_report.installed_tools, vec!["Dummy"]);
assert_eq!(feature_report.declared_background_tasks[0].name, "daily");
assert!(feature_report.declared_background_tasks.is_empty());
}
#[test]
@@ -2480,8 +2690,9 @@ mod tests {
})
.install_into_pending(&mut pending_tools, &mut hook_builder);
assert_eq!(pending_tools.len(), 1);
assert!(report.reports[0].installed);
assert!(pending_tools.is_empty());
assert!(!report.reports[0].installed);
assert!(report.reports[0].installed_tools.is_empty());
assert!(!report.reports[1].installed);
assert!(
report.reports[1]
@@ -2558,7 +2769,7 @@ mod tests {
"1.0.0",
"startup-discovered service",
))
.with_background_task(BackgroundTaskDeclaration::descriptor_only(
.with_background_task(BackgroundTaskDeclaration::worker_managed(
"provider-poller",
"provider lifecycle poller",
))
@@ -2568,7 +2779,7 @@ mod tests {
}
#[test]
fn protocol_provider_registers_startup_discovered_contributions_through_worker_path() {
fn protocol_provider_report_only_background_task_is_rejected_atomically() {
let provider = ProtocolProviderDeclaration::new(
ProviderId::builtin("dynamic-provider"),
"test-protocol",
@@ -2599,30 +2810,18 @@ mod tests {
.collect();
let feature_report = &report.reports[0];
assert!(feature_report.installed);
assert_eq!(feature_report.installed_tools, vec!["DynamicTool"]);
assert_eq!(tool_names, vec!["DynamicTool"]);
assert!(!feature_report.installed);
assert!(feature_report.installed_tools.is_empty());
assert!(tool_names.is_empty());
assert_eq!(calls.load(Ordering::SeqCst), 1);
assert_eq!(feature_report.provided_services.len(), 1);
assert_eq!(
feature_report.provided_services[0].id,
ServiceId::builtin("dynamic-service")
);
assert_eq!(
feature_report.declared_background_tasks[0].name,
"provider-poller"
);
assert!(feature_report.provided_services.is_empty());
assert!(feature_report.declared_background_tasks.is_empty());
assert_eq!(feature_report.protocol_providers.len(), 1);
assert_eq!(
feature_report.protocol_providers[0].state,
ProtocolProviderLifecycleState::Ready
);
assert!(
feature_report
.diagnostics
.iter()
.any(|diagnostic| diagnostic.message.contains("startup discovery completed"))
);
assert!(feature_report.diagnostics.iter().any(|diagnostic| {
diagnostic
.message
.contains("has no executable Worker-managed handler")
}));
}
#[test]
@@ -2779,8 +2978,8 @@ mod tests {
async fn call(
&self,
_input: &crate::hook::ToolCallSummary,
) -> crate::hook::HookPreToolAction {
crate::hook::HookPreToolAction::Continue
) -> Result<crate::hook::HookPreToolAction, crate::hook::HookError> {
Ok(crate::hook::HookPreToolAction::Continue)
}
}
@@ -2804,6 +3003,19 @@ mod tests {
}
}
struct NoopBackgroundTask;
#[async_trait]
impl FeatureBackgroundTask for NoopBackgroundTask {
async fn run(
&self,
_context: background::BackgroundTaskContext,
_cancellation: background::BackgroundTaskCancellation,
) -> Result<(), crate::hook::HookError> {
Ok(())
}
}
struct BackgroundFeature {
descriptor: FeatureDescriptor,
task_name: &'static str,
@@ -2818,12 +3030,22 @@ mod tests {
&self,
context: &mut FeatureInstallContext<'_>,
) -> Result<(), FeatureInstallError> {
context
.background_tasks()
.declare(BackgroundTaskDeclaration::descriptor_only(
self.task_name,
"runtime background task",
))
let declaration = self
.descriptor
.background_tasks
.iter()
.find(|task| task.name == self.task_name)
.cloned()
.unwrap_or_else(|| {
BackgroundTaskDeclaration::worker_managed(
self.task_name,
"undeclared background task",
)
});
context.background_tasks().register(
BackgroundTaskSpec::single_flight(declaration, std::time::Duration::from_secs(1)),
NoopBackgroundTask,
)
}
}
@@ -2985,25 +3207,44 @@ mod tests {
);
}
#[test]
fn background_task_declaration_is_descriptor_contribution() {
#[tokio::test]
async fn executable_background_task_is_registered_in_worker_scope() {
let descriptor = FeatureDescriptor::builtin("background", "Background")
.with_background_task(BackgroundTaskDeclaration::descriptor_only(
.with_background_task(BackgroundTaskDeclaration::worker_managed(
"declared-task",
"descriptor contribution",
));
let mut hook_builder = HookRegistryBuilder::default();
let mut pending_tools = Vec::new();
let report = FeatureRegistryBuilder::new()
.with_module(ServiceFeature { descriptor })
.with_module(BackgroundFeature {
descriptor,
task_name: "declared-task",
})
.install_into_pending(&mut pending_tools, &mut hook_builder);
assert!(report.reports[0].installed);
assert_eq!(
report.reports[0].declared_background_tasks[0].name,
"declared-task"
);
assert!(report.reports[0].skipped.is_empty());
assert!(matches!(
report
.background_tasks
.start(
&FeatureId::builtin("background"),
"declared-task",
crate::hook::HookInvocationContext::default(),
)
.unwrap(),
background::BackgroundTaskStart::Started { .. }
));
report.background_tasks.shutdown().await.unwrap();
assert!(matches!(
report.background_tasks.diagnostics()[0].outcome,
background::BackgroundTaskOutcome::Completed
| background::BackgroundTaskOutcome::Cancelled
));
}
#[test]
@@ -3118,7 +3359,10 @@ mod tests {
assert_eq!(descriptor.runtime, FeatureRuntimeKind::Builtin);
assert_eq!(
hook_points,
vec![FeatureHookPoint::PreRequest, FeatureHookPoint::PreToolCall]
vec![
FeatureHookPoint::PreLlmRequest,
FeatureHookPoint::PreToolCall
]
);
assert!(descriptor.background_tasks.is_empty());
assert!(descriptor.provides_services.is_empty());
File diff suppressed because it is too large Load Diff
+2 -3
View File
@@ -8,7 +8,8 @@ pub mod flow_transition;
pub mod manage_workdir;
pub mod manage_worker;
pub mod memory;
pub mod memory_extract;
pub(crate) mod memory_lifecycle;
pub mod memory_staging_output;
pub mod merge_request;
pub mod objective;
pub mod orchestration;
@@ -19,8 +20,6 @@ pub mod ticket;
pub mod worker_observation;
pub mod workspace_worker_discovery;
pub(crate) use memory_extract::{MemoryExtractFeature, MemoryExtractState, render_extract_input};
pub(crate) use session_explore::{SessionExploreFeature, SessionExploreState};
pub use task::{TaskFeature, task_tools_feature};
pub use ticket::{
TicketFeature, TicketFeatureAccess, ticket_tools_feature, ticket_tools_feature_with_access,
+308
View File
@@ -18,6 +18,10 @@ use schemars::JsonSchema;
use serde::de::DeserializeOwned;
use serde_json::json;
use crate::feature::{
FeatureDescriptor, FeatureInstallContext, FeatureInstallError, FeatureModule, ToolContribution,
ToolDeclaration,
};
use crate::worker::{
WorkspaceClient, WorkspaceClientError, WorkspaceRequest, WorkspaceRequestMethod,
};
@@ -338,6 +342,151 @@ fn query_schema() -> serde_json::Value {
})
}
pub struct MemoryFeatureInstallPlan {
pub module: MemoryToolsFeature,
pub resident_summary: Option<String>,
pub system_prompt_override: Option<String>,
pub(crate) resolved_config: manifest::ResolvedMemoryFeatureConfig,
}
impl MemoryFeatureInstallPlan {
pub async fn prepare(
manifest: &manifest::WorkerManifest,
client: Arc<dyn WorkspaceClient>,
prompts: Arc<crate::prompt::catalog::PromptCatalog>,
) -> std::io::Result<Option<Self>> {
Self::prepare_resolved(
manifest.feature.memory.clone(),
client,
prompts,
manifest.profile.clone(),
)
.await
}
async fn prepare_resolved(
config: manifest::ResolvedMemoryFeatureConfig,
client: Arc<dyn WorkspaceClient>,
prompts: Arc<crate::prompt::catalog::PromptCatalog>,
profile: Option<manifest::ProfileManifestSnapshot>,
) -> std::io::Result<Option<Self>> {
let memory_consolidation_worker = profile.as_ref().is_some_and(|snapshot| {
matches!(
&snapshot.source,
manifest::ProfileSource::Registry {
source: manifest::ProfileRegistrySource::Builtin,
name,
..
} if name == "memory-consolidation"
)
});
config
.validate_execution()
.map_err(|message| std::io::Error::new(std::io::ErrorKind::InvalidInput, message))?;
if !config.profile.enabled {
return Ok(None);
}
let workspace_id = client.workspace_id().ok_or_else(|| {
std::io::Error::new(
std::io::ErrorKind::InvalidInput,
"Memory tools require Backend Workspace API authority",
)
})?;
let settings = config
.workspace_settings()
.expect("validated enabled Memory config has Workspace settings");
if settings.workspace_id != workspace_id {
return Err(std::io::Error::new(
std::io::ErrorKind::InvalidInput,
format!(
"Memory settings belong to {} instead of {}",
settings.workspace_id, workspace_id
),
));
}
let resident_summary = if config.profile.resident.inject_summary {
match client
.execute_memory_backend_operation(
memory::backend::MemoryBackendOperation::ResidentSummary(
memory::backend::MemoryResidentSummaryOperation::default(),
),
)
.await
{
Ok(memory::backend::MemoryBackendOperationResult::ToolOutput(output)) => {
output.content
}
Ok(other) => {
tracing::debug!(?other, "unexpected resident Memory Backend result");
None
}
Err(error) => {
tracing::debug!(%error, "resident Memory summary unavailable");
None
}
}
} else {
None
};
let system_prompt_override = if memory_consolidation_worker {
let language = settings.language;
Some(
prompts
.memory_consolidation_system(&language)
.map_err(|error| std::io::Error::other(error.to_string()))?,
)
} else {
None
};
Ok(Some(Self {
module: MemoryToolsFeature::new(client, config.profile.staging_tools),
resident_summary,
system_prompt_override,
resolved_config: config,
}))
}
}
#[derive(Clone)]
pub struct MemoryToolsFeature {
tools: Vec<ToolDefinition>,
}
impl MemoryToolsFeature {
pub fn new(client: Arc<dyn WorkspaceClient>, staging_tools: bool) -> Self {
let tools = if staging_tools {
workspace_http_memory_consolidation_tools(client)
} else {
workspace_http_memory_tools(client)
};
Self { tools }
}
}
impl FeatureModule for MemoryToolsFeature {
fn descriptor(&self) -> FeatureDescriptor {
let mut descriptor = FeatureDescriptor::builtin("memory", "Memory")
.with_description("Workspace Memory document, query, and staging tools.");
for tool in &self.tools {
let (meta, _) = tool();
descriptor = descriptor.with_tool(ToolDeclaration::new(meta.name, meta.description));
}
descriptor
}
fn install(&self, context: &mut FeatureInstallContext<'_>) -> Result<(), FeatureInstallError> {
for tool in &self.tools {
let (meta, _) = tool();
context
.tools()
.register(ToolContribution::new(meta.name, tool.clone()))?;
}
Ok(())
}
}
#[cfg(test)]
mod tests {
use super::*;
@@ -350,6 +499,39 @@ mod tests {
))
}
fn resident_client(content: &str) -> Arc<dyn WorkspaceClient> {
use std::io::{Read, Write};
use std::net::TcpListener;
let listener = TcpListener::bind("127.0.0.1:0").unwrap();
let addr = listener.local_addr().unwrap();
let content = content.to_string();
std::thread::spawn(move || {
let (mut stream, _) = listener.accept().unwrap();
let mut request = [0_u8; 1024];
let _ = stream.read(&mut request).unwrap();
let body = serde_json::json!({
"status": "ok",
"result": {
"kind": "tool_output",
"summary": "resident Memory summary collected",
"content": content,
}
})
.to_string();
let response = format!(
"HTTP/1.1 200 OK\r\ncontent-type: application/json\r\ncontent-length: {}\r\nconnection: close\r\n\r\n{}",
body.len(),
body
);
stream.write_all(response.as_bytes()).unwrap();
});
Arc::new(crate::worker::TestWorkspaceHttpClient::new(
"workspace",
format!("http://{addr}"),
))
}
fn tool_names(definitions: Vec<ToolDefinition>) -> Vec<String> {
let mut names = definitions
.into_iter()
@@ -368,6 +550,132 @@ mod tests {
.input_schema
}
#[tokio::test]
async fn memory_install_plan_is_the_fail_closed_config_boundary() {
let prompts = crate::prompt::catalog::PromptCatalog::builtins_only().unwrap();
let disabled = MemoryFeatureInstallPlan::prepare_resolved(
manifest::ResolvedMemoryFeatureConfig::default(),
test_client(),
prompts.clone(),
None,
)
.await
.unwrap();
assert!(disabled.is_none());
let mut enabled = manifest::ResolvedMemoryFeatureConfig::default();
enabled.profile.enabled = true;
enabled.profile.resident.inject_summary = false;
assert!(
MemoryFeatureInstallPlan::prepare_resolved(
enabled.clone(),
test_client(),
prompts.clone(),
None,
)
.await
.is_err()
);
enabled
.bind_workspace_settings(manifest::WorkspaceMemorySettingsSnapshot {
workspace_id: "workspace".to_string(),
settings_revision: 1,
language: "English".to_string(),
})
.unwrap();
let mut foreign = enabled.clone();
foreign.workspace_settings.as_mut().unwrap().workspace_id = "other-workspace".to_string();
assert!(
MemoryFeatureInstallPlan::prepare_resolved(
foreign,
test_client(),
prompts.clone(),
None,
)
.await
.is_err()
);
let plan = MemoryFeatureInstallPlan::prepare_resolved(
enabled.clone(),
test_client(),
prompts.clone(),
None,
)
.await
.unwrap()
.unwrap();
assert!(plan.resident_summary.is_none());
assert!(plan.system_prompt_override.is_none());
enabled.profile.resident.inject_summary = true;
let plan = MemoryFeatureInstallPlan::prepare_resolved(
enabled,
resident_client("# Durable Memory"),
prompts,
None,
)
.await
.unwrap()
.unwrap();
assert_eq!(plan.resident_summary.as_deref(), Some("# Durable Memory"));
}
#[tokio::test]
async fn memory_prompt_contribution_rereads_resident_summary_for_each_install() {
let prompts = crate::prompt::catalog::PromptCatalog::builtins_only().unwrap();
let mut config = manifest::ResolvedMemoryFeatureConfig::default();
config.profile.enabled = true;
config
.bind_workspace_settings(manifest::WorkspaceMemorySettingsSnapshot {
workspace_id: "workspace".to_string(),
settings_revision: 1,
language: "English".to_string(),
})
.unwrap();
let first = MemoryFeatureInstallPlan::prepare_resolved(
config.clone(),
resident_client("first resident summary"),
prompts.clone(),
None,
)
.await
.unwrap()
.unwrap();
let restored = MemoryFeatureInstallPlan::prepare_resolved(
config,
resident_client("updated resident summary"),
prompts,
None,
)
.await
.unwrap()
.unwrap();
assert_eq!(
first.resident_summary.as_deref(),
Some("first resident summary")
);
assert_eq!(
restored.resident_summary.as_deref(),
Some("updated resident summary")
);
}
#[test]
fn memory_feature_owns_normal_and_staging_tool_surfaces() {
let normal = MemoryToolsFeature::new(test_client(), false);
let normal_names = tool_names(normal.tools);
assert!(normal_names.contains(&"MemoryQuery".to_string()));
assert!(!normal_names.contains(&"MemoryStagingList".to_string()));
let staging = MemoryToolsFeature::new(test_client(), true);
assert_eq!(staging.descriptor().id.as_str(), "builtin:memory");
let staging_names = tool_names(staging.tools);
assert!(staging_names.contains(&"MemoryQuery".to_string()));
assert!(staging_names.contains(&"MemoryStagingList".to_string()));
}
#[test]
fn normal_workspace_memory_tools_do_not_include_staging_tools() {
let names = tool_names(workspace_http_memory_tools(test_client()));
File diff suppressed because it is too large Load Diff
@@ -28,7 +28,7 @@ const FINISH_DESCRIPTION: &str =
"Finish Memory extraction after validating the number of candidates staged during this run.";
#[derive(Clone)]
pub(crate) struct MemoryExtractState {
pub(crate) struct MemoryStagingOutputState {
view: Arc<SessionCapture>,
workspace_client: Arc<dyn WorkspaceClient>,
source: SourceRef,
@@ -37,7 +37,7 @@ pub(crate) struct MemoryExtractState {
finished: Arc<Mutex<Option<FinishMemoryExtractionParams>>>,
}
impl MemoryExtractState {
impl MemoryStagingOutputState {
pub(crate) fn new(
view: SessionCapture,
workspace_client: Arc<dyn WorkspaceClient>,
@@ -70,22 +70,20 @@ impl MemoryExtractState {
}
#[derive(Clone)]
pub(crate) struct MemoryExtractFeature {
state: MemoryExtractState,
pub(crate) struct MemoryStagingOutputFeature {
state: MemoryStagingOutputState,
}
impl MemoryExtractFeature {
pub(crate) fn new(state: MemoryExtractState) -> Self {
impl MemoryStagingOutputFeature {
pub(crate) fn new(state: MemoryStagingOutputState) -> Self {
Self { state }
}
}
impl FeatureModule for MemoryExtractFeature {
impl FeatureModule for MemoryStagingOutputFeature {
fn descriptor(&self) -> FeatureDescriptor {
FeatureDescriptor::builtin("memory-extract", "Memory Extract")
.with_description(
"Memory staging and extraction completion, independent from session exploration.",
)
FeatureDescriptor::builtin("memory-staging-output", "Memory Staging Output")
.with_description("Restricted Memory staging output for an extraction Internal Worker.")
.with_tool(ToolDeclaration::new(
"StageMemoryCandidate",
STAGE_DESCRIPTION,
@@ -109,7 +107,7 @@ impl FeatureModule for MemoryExtractFeature {
}
}
fn stage_definition(state: MemoryExtractState) -> ToolDefinition {
fn stage_definition(state: MemoryStagingOutputState) -> ToolDefinition {
Arc::new(move || {
let schema = serde_json::to_value(schemars::schema_for!(StageMemoryCandidateParams))
.unwrap_or_else(|_| serde_json::json!({}));
@@ -123,7 +121,7 @@ fn stage_definition(state: MemoryExtractState) -> ToolDefinition {
})
}
fn finish_definition(state: MemoryExtractState) -> ToolDefinition {
fn finish_definition(state: MemoryStagingOutputState) -> ToolDefinition {
Arc::new(move || {
let schema = serde_json::to_value(schemars::schema_for!(FinishMemoryExtractionParams))
.unwrap_or_else(|_| serde_json::json!({}));
@@ -157,7 +155,7 @@ struct FinishMemoryExtractionParams {
}
struct StageMemoryCandidateTool {
state: MemoryExtractState,
state: MemoryStagingOutputState,
}
#[async_trait]
@@ -252,7 +250,7 @@ impl Tool for StageMemoryCandidateTool {
}
struct FinishMemoryExtractionTool {
state: MemoryExtractState,
state: MemoryStagingOutputState,
}
#[async_trait]
@@ -431,8 +429,8 @@ mod tests {
use super::*;
fn state() -> MemoryExtractState {
MemoryExtractState::new(
fn state() -> MemoryStagingOutputState {
MemoryStagingOutputState::new(
SessionCapture::new("segment-1", vec![Item::user_message("durable decision")]),
crate::worker::marker_workspace_client(None, "test-backend"),
SourceRef {
@@ -445,8 +443,8 @@ mod tests {
#[test]
fn memory_extract_declares_only_memory_mutation_tools() {
let descriptor = MemoryExtractFeature::new(state()).descriptor();
assert_eq!(descriptor.id.as_str(), "builtin:memory-extract");
let descriptor = MemoryStagingOutputFeature::new(state()).descriptor();
assert_eq!(descriptor.id.as_str(), "builtin:memory-staging-output");
assert_eq!(
descriptor
.tools
+13 -7
View File
@@ -125,7 +125,7 @@ impl FeatureModule for TaskFeature {
))
.with_hook(HookDeclaration::new(
"task-reminder-pre-request",
FeatureHookPoint::PreRequest,
FeatureHookPoint::PreLlmRequest,
))
.with_hook(HookDeclaration::new(
"task-reminder-tool-usage",
@@ -209,24 +209,27 @@ struct TaskReminderPreRequestHook {
#[async_trait]
impl Hook<PreLlmRequest> for TaskReminderPreRequestHook {
async fn call(&self, input: &PreRequestContext) -> HookPreRequestAction {
async fn call(
&self,
input: &PreRequestContext,
) -> Result<HookPreRequestAction, crate::hook::HookError> {
let tasks = self.state.task_store.list();
if tasks.is_empty() {
return HookPreRequestAction::Continue;
return Ok(HookPreRequestAction::Continue);
}
let (since_task_management, since_reminder) = self.state.reminder_state.note_request();
if since_task_management < TASK_REMINDER_REQUEST_THRESHOLD
|| since_reminder < TASK_REMINDER_COOLDOWN_REQUESTS
{
return HookPreRequestAction::Continue;
return Ok(HookPreRequestAction::Continue);
}
if let Some(system_items) = input.system_items() {
self.state.reminder_state.note_reminder();
system_items.append_task_reminder(render_task_reminder_body(&tasks));
}
HookPreRequestAction::Continue
Ok(HookPreRequestAction::Continue)
}
}
@@ -236,11 +239,14 @@ struct TaskReminderToolUsageHook {
#[async_trait]
impl Hook<PreToolCall> for TaskReminderToolUsageHook {
async fn call(&self, input: &ToolCallSummary) -> HookPreToolAction {
async fn call(
&self,
input: &ToolCallSummary,
) -> Result<HookPreToolAction, crate::hook::HookError> {
if is_task_management_tool(&input.tool_name) {
self.state.reminder_state.note_task_management();
}
HookPreToolAction::Continue
Ok(HookPreToolAction::Continue)
}
}
+1 -2
View File
@@ -2115,8 +2115,7 @@ read exit_notification || true
meta.name
})
.collect();
assert!(!names.iter().any(|name| name == "Mcp_demo_search_files"));
assert!(names.iter().any(|name| name == "Mcp_demo_unique"));
assert!(names.is_empty());
}
fn shell_tool_server(response: &str) -> McpStdioServerSpec {
+1 -1
View File
@@ -7621,7 +7621,7 @@ mod tests {
)))
.install_into_pending(&mut pending, &mut hooks);
assert_eq!(pending.len(), 1);
assert!(pending.is_empty());
assert_eq!(skipped_count(&report), 1);
assert!(has_diagnostic(&report, "duplicate tool contribution"));
}
+119
View File
@@ -0,0 +1,119 @@
use std::sync::Arc;
use agen::{HistoryEntry, UsageRecord};
use serde_json::Value;
use crate::session_history::SessionHistoryMetadata;
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
pub(crate) enum CommittedRunExit {
Finished,
NonFinal,
Interrupted,
}
/// Immutable projection of one durably committed session-log location.
///
/// Feature code receives this value only after the host has committed the
/// terminal run record. The projection deliberately carries annotated history
/// rather than the public flattened transcript so provenance-sensitive
/// features can construct their own bounded views.
#[derive(Clone)]
pub(crate) struct CommittedSessionCapture {
pub(crate) session_id: String,
pub(crate) segment_id: String,
/// Monotonic committed-log revision for the captured Segment.
pub(crate) session_revision: u64,
pub(crate) entry_count: usize,
pub(crate) run_exit: CommittedRunExit,
pub(crate) history: Vec<HistoryEntry<SessionHistoryMetadata>>,
pub(crate) usage_history: Vec<UsageRecord>,
pub(crate) extensions: Vec<(String, Value)>,
}
#[derive(Clone, Debug, PartialEq, Eq)]
pub(crate) struct CommittedSessionLocation {
pub(crate) session_id: String,
pub(crate) segment_id: String,
/// Monotonic committed-log revision for the captured Segment.
pub(crate) session_revision: u64,
pub(crate) entry_count: usize,
}
impl CommittedSessionCapture {
pub(crate) fn location(&self) -> CommittedSessionLocation {
CommittedSessionLocation {
session_id: self.session_id.clone(),
segment_id: self.segment_id.clone(),
session_revision: self.session_revision,
entry_count: self.entry_count,
}
}
}
#[derive(Debug, thiserror::Error)]
pub(crate) enum FeatureSessionError {
#[error("read committed session failed: {0}")]
Capture(String),
#[error("append session extension failed: {0}")]
Extension(String),
}
#[derive(Clone)]
pub(crate) struct CommittedSessionCaptureHandle {
capture: Arc<
dyn Fn() -> Result<CommittedSessionCapture, FeatureSessionError> + Send + Sync + 'static,
>,
}
impl CommittedSessionCaptureHandle {
pub(crate) fn new(
capture: impl Fn() -> Result<CommittedSessionCapture, FeatureSessionError>
+ Send
+ Sync
+ 'static,
) -> Self {
Self {
capture: Arc::new(capture),
}
}
pub(crate) fn capture(&self) -> Result<CommittedSessionCapture, FeatureSessionError> {
(self.capture)()
}
}
#[derive(Clone)]
pub(crate) struct SessionExtensionHandle {
append: Arc<
dyn Fn(&CommittedSessionLocation, &str, Value) -> Result<bool, FeatureSessionError>
+ Send
+ Sync
+ 'static,
>,
}
impl SessionExtensionHandle {
pub(crate) fn new(
append: impl Fn(&CommittedSessionLocation, &str, Value) -> Result<bool, FeatureSessionError>
+ Send
+ Sync
+ 'static,
) -> Self {
Self {
append: Arc::new(append),
}
}
/// Appends an extension only while the committed session is still at the
/// exact location captured by the feature. `Ok(false)` is a stale-write
/// fence, not an I/O failure.
pub(crate) fn append_if_current(
&self,
expected: &CommittedSessionLocation,
domain: &str,
payload: Value,
) -> Result<bool, FeatureSessionError> {
(self.append)(expected, domain, payload)
}
}
+645 -52
View File
@@ -23,8 +23,105 @@ use agen::interceptor::{
};
use agen::tool::{ToolOutput, ToolResult};
use async_trait::async_trait;
use serde::{Deserialize, Serialize};
use serde_json::Value;
use session_store::{SystemItem, SystemReminder};
use thiserror::Error;
use crate::SessionEntryRef;
const HOOK_DIAGNOSTIC_MAX_BYTES: usize = 1_024;
/// Failure category exposed by the safe Worker hook boundary.
///
/// Categories are intentionally closed and payload-free so extensions cannot
/// smuggle provider, credential, or history data into diagnostics.
#[derive(Clone, Copy, Debug, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "snake_case")]
pub enum HookErrorCategory {
InvalidInput,
Dependency,
Timeout,
Cancelled,
Trap,
ScopeDisposed,
Internal,
}
/// Bounded hook callback failure. Raw tool arguments, output, prompts, and
/// credentials must never be placed in `diagnostic`.
#[derive(Clone, Debug, PartialEq, Eq, Error)]
#[error("{category:?}: {diagnostic}")]
pub struct HookError {
pub category: HookErrorCategory,
pub diagnostic: String,
}
impl HookError {
pub fn new(category: HookErrorCategory, diagnostic: impl Into<String>) -> Self {
Self {
category,
diagnostic: bounded_utf8(diagnostic.into(), HOOK_DIAGNOSTIC_MAX_BYTES),
}
}
}
/// Failure behavior declared when a hook is registered.
#[derive(Clone, Copy, Debug, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "snake_case")]
pub enum HookFailurePolicy {
/// Gate the current operation when the hook cannot decide safely.
FailClosed,
/// Keep the already-authorized operation moving and emit a diagnostic.
FailOpenWithDiagnostic,
/// Keep committed state intact and mark the failure for operator attention.
AttentionRequired,
}
const DEFAULT_HOOK_TIMEOUT_MS: u64 = 30_000;
const MAX_HOOK_TIMEOUT_MS: u64 = 120_000;
/// Host-enforced execution and failure budget for one hook registration.
#[derive(Clone, Copy, Debug, PartialEq, Eq, Serialize, Deserialize)]
pub struct HookExecutionPolicy {
pub failure: HookFailurePolicy,
pub timeout_ms: u64,
}
impl HookExecutionPolicy {
pub const fn new(failure: HookFailurePolicy, timeout_ms: u64) -> Self {
Self {
failure,
timeout_ms,
}
}
pub const fn fail_closed() -> Self {
Self::new(HookFailurePolicy::FailClosed, DEFAULT_HOOK_TIMEOUT_MS)
}
fn validate(self) -> Result<Self, HookError> {
if self.timeout_ms == 0 || self.timeout_ms > MAX_HOOK_TIMEOUT_MS {
return Err(HookError::new(
HookErrorCategory::InvalidInput,
format!("hook timeout_ms must be within 1..={MAX_HOOK_TIMEOUT_MS}"),
));
}
Ok(self)
}
}
fn bounded_utf8(mut value: String, max_bytes: usize) -> String {
if value.len() <= max_bytes {
return value;
}
let mut end = max_bytes;
while end > 0 && !value.is_char_boundary(end) {
end -= 1;
}
value.truncate(end);
value
}
/// Hook-facing prompt-submit action.
///
@@ -285,12 +382,6 @@ pub struct TurnEndInfo {
pub final_text_preview: String,
}
/// Information passed to `OnAbort` hooks.
pub struct AbortInfo {
/// Reason supplied by the aborter.
pub reason: String,
}
// =============================================================================
// Hook Event Kinds
// =============================================================================
@@ -315,10 +406,8 @@ pub struct PreLlmRequest;
pub struct PreToolCall;
/// After each tool completes; observational except it may abort the run.
pub struct PostToolCall;
/// When a turn ends with no tool calls; observational except it may pause.
/// After every terminal assistant response is committed; observational except it may pause.
pub struct OnTurnEnd;
/// When execution is interrupted; observational only.
pub struct OnAbort;
impl HookEventKind for OnPromptSubmit {
type Input = PromptSubmitInfo;
@@ -345,11 +434,6 @@ impl HookEventKind for OnTurnEnd {
type Output = HookTurnEndAction;
}
impl HookEventKind for OnAbort {
type Input = AbortInfo;
type Output = ();
}
// =============================================================================
// Hook Trait
// =============================================================================
@@ -362,25 +446,194 @@ impl HookEventKind for OnAbort {
/// short-circuit on the first non-continue action.
#[async_trait]
pub trait Hook<E: HookEventKind>: Send + Sync {
async fn call(&self, input: &E::Input) -> E::Output;
async fn call(&self, input: &E::Input) -> Result<E::Output, HookError>;
}
// =============================================================================
// Hook Registry
// =============================================================================
/// Bounded identity-only view of the committed history visible at a lifecycle
/// boundary. Hook context never carries history payloads; a Feature that needs
/// content must use a separately granted session-exploration service.
#[derive(Clone, Debug, Default, PartialEq, Eq)]
pub struct HookHistoryRange {
pub first_entry_ref: Option<SessionEntryRef>,
pub last_entry_ref: Option<SessionEntryRef>,
pub entry_count: usize,
}
/// Stable provenance attached to every Worker lifecycle callback.
#[derive(Clone, Debug, Default, PartialEq, Eq)]
pub struct HookInvocationContext {
pub workspace_id: Option<String>,
pub worker_id: String,
pub session_id: String,
pub session_revision: u64,
pub run_id: Option<String>,
pub turn_index: Option<usize>,
pub call_id: Option<String>,
}
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
pub enum RunCommittedExit {
Finished,
Paused,
Yielded,
Interrupted,
}
#[derive(Clone, Debug, PartialEq, Eq)]
pub struct RunExitContext {
pub invocation: HookInvocationContext,
pub exit: RunCommittedExit,
pub history_len: usize,
}
#[derive(Clone, Debug, PartialEq)]
pub struct RunCommittedContext {
pub invocation: HookInvocationContext,
pub exit: RunCommittedExit,
pub committed_history: HookHistoryRange,
}
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
pub enum SessionRewriteKind {
Rewind,
Compact,
Fork,
Restore,
}
#[derive(Clone, Debug, PartialEq)]
pub struct BeforeSessionRewriteContext {
pub invocation: HookInvocationContext,
pub kind: SessionRewriteKind,
pub current_history: HookHistoryRange,
}
#[derive(Clone, Debug, PartialEq, Eq)]
pub enum BeforeSessionRewriteAction {
Continue,
Deny(String),
}
#[derive(Clone, Debug, PartialEq, Eq)]
pub struct WorkerStoppingContext {
pub invocation: HookInvocationContext,
pub reason: String,
}
pub struct RunExit;
pub struct RunCommitted;
pub struct BeforeSessionRewrite;
pub struct WorkerStopping;
impl HookEventKind for RunExit {
type Input = RunExitContext;
type Output = ();
}
impl HookEventKind for RunCommitted {
type Input = RunCommittedContext;
type Output = ();
}
impl HookEventKind for BeforeSessionRewrite {
type Input = BeforeSessionRewriteContext;
type Output = BeforeSessionRewriteAction;
}
impl HookEventKind for WorkerStopping {
type Input = WorkerStoppingContext;
type Output = ();
}
pub(crate) struct RegisteredHook<E: HookEventKind> {
owner: String,
policy: HookExecutionPolicy,
hook: Box<dyn Hook<E>>,
}
impl<E: HookEventKind> RegisteredHook<E> {
pub(crate) async fn call(&self, input: &E::Input) -> Result<E::Output, HookExecutionError> {
let result = tokio::time::timeout(
std::time::Duration::from_millis(self.policy.timeout_ms),
self.hook.call(input),
)
.await
.unwrap_or_else(|_| {
Err(HookError::new(
HookErrorCategory::Timeout,
"hook exceeded its host-enforced execution deadline",
))
});
result.map_err(|source| HookExecutionError {
owner: self.owner.clone(),
policy: self.policy.failure,
source,
})
}
pub(crate) async fn call_optional(
&self,
input: &E::Input,
) -> Result<Option<E::Output>, HookExecutionError> {
match self.call(input).await {
Ok(output) => Ok(Some(output)),
Err(error) if error.policy == HookFailurePolicy::FailOpenWithDiagnostic => {
tracing::warn!(owner = %error.owner, error = %error.source, "inline hook failed open");
Ok(None)
}
Err(error) => Err(error),
}
}
}
#[derive(Clone, Debug, PartialEq, Eq, Error)]
#[error("hook `{owner}` failed under {policy:?}: {source}")]
pub struct HookExecutionError {
pub owner: String,
pub policy: HookFailurePolicy,
pub source: HookError,
}
/// Builder for constructing a frozen `HookRegistry`.
///
/// Hooks are added during setup, then `build()` produces an immutable
/// registry that can be shared via `Arc`.
#[derive(Default)]
pub struct HookRegistryBuilder {
on_prompt_submit: Vec<Box<dyn Hook<OnPromptSubmit>>>,
pre_llm_request: Vec<Box<dyn Hook<PreLlmRequest>>>,
pre_tool_call: Vec<Box<dyn Hook<PreToolCall>>>,
post_tool_call: Vec<Box<dyn Hook<PostToolCall>>>,
on_turn_end: Vec<Box<dyn Hook<OnTurnEnd>>>,
on_abort: Vec<Box<dyn Hook<OnAbort>>>,
on_prompt_submit: Vec<RegisteredHook<OnPromptSubmit>>,
pre_llm_request: Vec<RegisteredHook<PreLlmRequest>>,
pre_tool_call: Vec<RegisteredHook<PreToolCall>>,
post_tool_call: Vec<RegisteredHook<PostToolCall>>,
on_turn_end: Vec<RegisteredHook<OnTurnEnd>>,
run_exit: Vec<RegisteredHook<RunExit>>,
run_committed: Vec<RegisteredHook<RunCommitted>>,
before_session_rewrite: Vec<RegisteredHook<BeforeSessionRewrite>>,
worker_stopping: Vec<RegisteredHook<WorkerStopping>>,
}
macro_rules! add_hook_methods {
($default:ident, $named:ident, $field:ident, $event:ty) => {
pub fn $default(&mut self, hook: impl Hook<$event> + 'static) {
self.$named("worker.host", HookExecutionPolicy::fail_closed(), hook)
.expect("host hook policy is valid");
}
pub fn $named(
&mut self,
owner: impl Into<String>,
policy: HookExecutionPolicy,
hook: impl Hook<$event> + 'static,
) -> Result<(), HookError> {
let policy = policy.validate()?;
self.$field.push(RegisteredHook {
owner: owner.into(),
policy,
hook: Box::new(hook),
});
Ok(())
}
};
}
impl HookRegistryBuilder {
@@ -388,31 +641,82 @@ impl HookRegistryBuilder {
Self::default()
}
pub fn add_on_prompt_submit(&mut self, hook: impl Hook<OnPromptSubmit> + 'static) {
self.on_prompt_submit.push(Box::new(hook));
add_hook_methods!(
add_on_prompt_submit,
add_named_on_prompt_submit,
on_prompt_submit,
OnPromptSubmit
);
add_hook_methods!(
add_pre_llm_request,
add_named_pre_llm_request,
pre_llm_request,
PreLlmRequest
);
add_hook_methods!(
add_pre_tool_call,
add_named_pre_tool_call,
pre_tool_call,
PreToolCall
);
add_hook_methods!(
add_post_tool_call,
add_named_post_tool_call,
post_tool_call,
PostToolCall
);
add_hook_methods!(
add_on_turn_end,
add_named_on_turn_end,
on_turn_end,
OnTurnEnd
);
add_hook_methods!(add_run_exit, add_named_run_exit, run_exit, RunExit);
add_hook_methods!(
add_run_committed,
add_named_run_committed,
run_committed,
RunCommitted
);
add_hook_methods!(
add_before_session_rewrite,
add_named_before_session_rewrite,
before_session_rewrite,
BeforeSessionRewrite
);
add_hook_methods!(
add_worker_stopping,
add_named_worker_stopping,
worker_stopping,
WorkerStopping
);
pub(crate) fn checkpoint(&self) -> [usize; 9] {
[
self.on_prompt_submit.len(),
self.pre_llm_request.len(),
self.pre_tool_call.len(),
self.post_tool_call.len(),
self.on_turn_end.len(),
self.run_exit.len(),
self.run_committed.len(),
self.before_session_rewrite.len(),
self.worker_stopping.len(),
]
}
pub fn add_pre_llm_request(&mut self, hook: impl Hook<PreLlmRequest> + 'static) {
self.pre_llm_request.push(Box::new(hook));
pub(crate) fn rollback_to(&mut self, checkpoint: [usize; 9]) {
self.on_prompt_submit.truncate(checkpoint[0]);
self.pre_llm_request.truncate(checkpoint[1]);
self.pre_tool_call.truncate(checkpoint[2]);
self.post_tool_call.truncate(checkpoint[3]);
self.on_turn_end.truncate(checkpoint[4]);
self.run_exit.truncate(checkpoint[5]);
self.run_committed.truncate(checkpoint[6]);
self.before_session_rewrite.truncate(checkpoint[7]);
self.worker_stopping.truncate(checkpoint[8]);
}
pub fn add_pre_tool_call(&mut self, hook: impl Hook<PreToolCall> + 'static) {
self.pre_tool_call.push(Box::new(hook));
}
pub fn add_post_tool_call(&mut self, hook: impl Hook<PostToolCall> + 'static) {
self.post_tool_call.push(Box::new(hook));
}
pub fn add_on_turn_end(&mut self, hook: impl Hook<OnTurnEnd> + 'static) {
self.on_turn_end.push(Box::new(hook));
}
pub fn add_on_abort(&mut self, hook: impl Hook<OnAbort> + 'static) {
self.on_abort.push(Box::new(hook));
}
/// Freeze the builder into an immutable registry.
pub fn build(self) -> HookRegistry {
HookRegistry {
on_prompt_submit: self.on_prompt_submit,
@@ -420,19 +724,140 @@ impl HookRegistryBuilder {
pre_tool_call: self.pre_tool_call,
post_tool_call: self.post_tool_call,
on_turn_end: self.on_turn_end,
on_abort: self.on_abort,
run_exit: self.run_exit,
run_committed: self.run_committed,
before_session_rewrite: self.before_session_rewrite,
worker_stopping: self.worker_stopping,
diagnostics: std::sync::Mutex::new(Vec::new()),
}
}
}
/// Frozen registry of hooks. Constructed via [`HookRegistryBuilder::build()`].
pub struct HookRegistry {
pub(crate) on_prompt_submit: Vec<Box<dyn Hook<OnPromptSubmit>>>,
pub(crate) pre_llm_request: Vec<Box<dyn Hook<PreLlmRequest>>>,
pub(crate) pre_tool_call: Vec<Box<dyn Hook<PreToolCall>>>,
pub(crate) post_tool_call: Vec<Box<dyn Hook<PostToolCall>>>,
pub(crate) on_turn_end: Vec<Box<dyn Hook<OnTurnEnd>>>,
pub(crate) on_abort: Vec<Box<dyn Hook<OnAbort>>>,
pub(crate) on_prompt_submit: Vec<RegisteredHook<OnPromptSubmit>>,
pub(crate) pre_llm_request: Vec<RegisteredHook<PreLlmRequest>>,
pub(crate) pre_tool_call: Vec<RegisteredHook<PreToolCall>>,
pub(crate) post_tool_call: Vec<RegisteredHook<PostToolCall>>,
pub(crate) on_turn_end: Vec<RegisteredHook<OnTurnEnd>>,
run_exit: Vec<RegisteredHook<RunExit>>,
run_committed: Vec<RegisteredHook<RunCommitted>>,
before_session_rewrite: Vec<RegisteredHook<BeforeSessionRewrite>>,
worker_stopping: Vec<RegisteredHook<WorkerStopping>>,
diagnostics: std::sync::Mutex<Vec<HookExecutionError>>,
}
impl HookRegistry {
fn record_diagnostic(&self, error: HookExecutionError) {
let mut diagnostics = self.diagnostics.lock().expect("hook diagnostics poisoned");
diagnostics.push(error);
if diagnostics.len() > 128 {
let remove = diagnostics.len() - 128;
diagnostics.drain(..remove);
}
}
pub(crate) fn record_chain_timeout(&self, lifecycle: &str) {
self.record_diagnostic(HookExecutionError {
owner: format!("worker.{lifecycle}"),
policy: HookFailurePolicy::AttentionRequired,
source: HookError::new(
HookErrorCategory::Timeout,
"hook chain exceeded its host-enforced lifecycle deadline",
),
});
}
pub fn diagnostics(&self) -> Vec<HookExecutionError> {
self.diagnostics
.lock()
.expect("hook diagnostics poisoned")
.clone()
}
pub async fn on_run_exit(&self, context: &RunExitContext) -> Result<(), HookExecutionError> {
for registration in &self.run_exit {
if let Err(error) = registration.call(context).await {
self.record_diagnostic(error.clone());
match error.policy {
HookFailurePolicy::FailOpenWithDiagnostic => {
tracing::warn!(owner = %error.owner, error = %error.source, "run-exit hook failed open");
}
HookFailurePolicy::FailClosed | HookFailurePolicy::AttentionRequired => {
return Err(error);
}
}
}
}
Ok(())
}
pub async fn on_run_committed(
&self,
context: &RunCommittedContext,
) -> Result<(), HookExecutionError> {
for registration in &self.run_committed {
if let Err(error) = registration.call(context).await {
self.record_diagnostic(error.clone());
match error.policy {
HookFailurePolicy::FailOpenWithDiagnostic => {
tracing::warn!(owner = %error.owner, error = %error.source, "run-committed hook failed open");
}
HookFailurePolicy::FailClosed | HookFailurePolicy::AttentionRequired => {
return Err(error);
}
}
}
}
Ok(())
}
pub async fn before_session_rewrite(
&self,
context: &BeforeSessionRewriteContext,
) -> Result<BeforeSessionRewriteAction, HookExecutionError> {
let mut denials = Vec::new();
for registration in &self.before_session_rewrite {
match registration.call(context).await {
Ok(BeforeSessionRewriteAction::Continue) => {}
Ok(BeforeSessionRewriteAction::Deny(reason)) => {
denials.push((registration.owner.clone(), reason));
}
Err(error) if error.policy == HookFailurePolicy::FailOpenWithDiagnostic => {
self.record_diagnostic(error.clone());
tracing::warn!(owner = %error.owner, error = %error.source, "session-rewrite hook failed open");
}
Err(error) => {
self.record_diagnostic(error.clone());
return Err(error);
}
}
}
denials.sort_by(|left, right| left.0.cmp(&right.0));
Ok(denials
.into_iter()
.next()
.map(|(_, reason)| BeforeSessionRewriteAction::Deny(reason))
.unwrap_or(BeforeSessionRewriteAction::Continue))
}
pub async fn on_worker_stopping(
&self,
context: &WorkerStoppingContext,
) -> Result<(), HookExecutionError> {
for registration in &self.worker_stopping {
if let Err(error) = registration.call(context).await {
self.record_diagnostic(error.clone());
match error.policy {
HookFailurePolicy::FailOpenWithDiagnostic
| HookFailurePolicy::AttentionRequired => {
tracing::warn!(owner = %error.owner, error = %error.source, "worker-stopping hook requires attention");
}
HookFailurePolicy::FailClosed => return Err(error),
}
}
}
Ok(())
}
}
#[cfg(test)]
@@ -494,4 +919,172 @@ mod tests {
let pause_action = HookPreToolAction::Pause.into_worker_action("call_4".into());
assert!(matches!(pause_action, PreToolAction::Pause));
}
struct RewriteHook {
action: BeforeSessionRewriteAction,
}
#[async_trait]
impl Hook<BeforeSessionRewrite> for RewriteHook {
async fn call(
&self,
_input: &BeforeSessionRewriteContext,
) -> Result<BeforeSessionRewriteAction, HookError> {
Ok(self.action.clone())
}
}
struct FailingRewriteHook;
#[async_trait]
impl Hook<BeforeSessionRewrite> for FailingRewriteHook {
async fn call(
&self,
_input: &BeforeSessionRewriteContext,
) -> Result<BeforeSessionRewriteAction, HookError> {
Err(HookError::new(
HookErrorCategory::Dependency,
"provider unavailable",
))
}
}
fn rewrite_context() -> BeforeSessionRewriteContext {
BeforeSessionRewriteContext {
invocation: HookInvocationContext {
workspace_id: Some("workspace".into()),
worker_id: "worker".into(),
session_id: "session".into(),
session_revision: 4,
run_id: None,
turn_index: None,
call_id: None,
},
kind: SessionRewriteKind::Compact,
current_history: HookHistoryRange::default(),
}
}
#[tokio::test]
async fn rewrite_denials_are_resolved_by_owner_not_registration_order() {
let mut builder = HookRegistryBuilder::new();
builder
.add_named_before_session_rewrite(
"z-feature",
HookExecutionPolicy::fail_closed(),
RewriteHook {
action: BeforeSessionRewriteAction::Deny("z denied".into()),
},
)
.unwrap();
builder
.add_named_before_session_rewrite(
"a-feature",
HookExecutionPolicy::fail_closed(),
RewriteHook {
action: BeforeSessionRewriteAction::Deny("a denied".into()),
},
)
.unwrap();
assert_eq!(
builder
.build()
.before_session_rewrite(&rewrite_context())
.await
.unwrap(),
BeforeSessionRewriteAction::Deny("a denied".into())
);
}
#[tokio::test]
async fn hook_failure_policy_is_applied_at_the_registry_boundary() {
let mut fail_open = HookRegistryBuilder::new();
fail_open
.add_named_before_session_rewrite(
"feature",
HookExecutionPolicy::new(HookFailurePolicy::FailOpenWithDiagnostic, 30_000),
FailingRewriteHook,
)
.unwrap();
let fail_open = fail_open.build();
assert_eq!(
fail_open
.before_session_rewrite(&rewrite_context())
.await
.unwrap(),
BeforeSessionRewriteAction::Continue
);
assert_eq!(fail_open.diagnostics().len(), 1);
let mut fail_closed = HookRegistryBuilder::new();
fail_closed
.add_named_before_session_rewrite(
"feature",
HookExecutionPolicy::fail_closed(),
FailingRewriteHook,
)
.unwrap();
let error = fail_closed
.build()
.before_session_rewrite(&rewrite_context())
.await
.unwrap_err();
assert_eq!(error.source.category, HookErrorCategory::Dependency);
}
struct NeverReturns;
#[async_trait]
impl Hook<BeforeSessionRewrite> for NeverReturns {
async fn call(
&self,
_input: &BeforeSessionRewriteContext,
) -> Result<BeforeSessionRewriteAction, HookError> {
std::future::pending().await
}
}
#[tokio::test]
async fn hook_execution_deadline_cancels_a_non_cooperative_callback() {
let mut builder = HookRegistryBuilder::new();
builder
.add_named_before_session_rewrite(
"feature",
HookExecutionPolicy::new(HookFailurePolicy::FailClosed, 10),
NeverReturns,
)
.unwrap();
let registry = builder.build();
let error = tokio::time::timeout(
std::time::Duration::from_millis(100),
registry.before_session_rewrite(&rewrite_context()),
)
.await
.expect("host hook deadline must terminate the callback")
.unwrap_err();
assert_eq!(error.source.category, HookErrorCategory::Timeout);
assert_eq!(registry.diagnostics().len(), 1);
}
#[test]
fn invalid_hook_execution_budget_is_rejected_before_registration() {
let mut builder = HookRegistryBuilder::new();
let error = builder
.add_named_before_session_rewrite(
"feature",
HookExecutionPolicy::new(HookFailurePolicy::FailClosed, 0),
NeverReturns,
)
.unwrap_err();
assert_eq!(error.category, HookErrorCategory::InvalidInput);
}
#[test]
fn hook_diagnostics_are_utf8_bounded() {
let error = HookError::new(HookErrorCategory::Internal, "".repeat(1_000));
assert!(error.diagnostic.len() <= HOOK_DIAGNOSTIC_MAX_BYTES);
assert!(error.diagnostic.is_char_boundary(error.diagnostic.len()));
}
}
+35 -29
View File
@@ -123,14 +123,14 @@ where
// Internal identities are run-scoped and never enter the public Runtime Worker catalog.
manifest.worker.name = format!("internal-{}-{}", identity.kind, identity.run_id);
// Internal jobs only receive features supplied below. A parent manifest must not accidentally
// grant its normal public tool surface or recursively schedule memory work.
// Internal jobs only receive the explicitly supplied Feature set below. A
// parent manifest cannot accidentally grant its normal public tool surface
// or recursively schedule Feature-owned background work.
manifest.feature = Default::default();
manifest.plugins = Default::default();
manifest.mcp = Default::default();
manifest.skills = None;
manifest.compaction = None;
manifest.memory = None;
let last_usage = Arc::new(Mutex::new(None::<UsageEvent>));
let usage_slot = last_usage.clone();
@@ -164,6 +164,7 @@ where
identity: identity.clone(),
history_entries: 0,
})?;
worker.disable_manifest_lifecycle_features();
if let Some(session) = inherited_workdir_session {
worker.bind_workdir_session(Some(session));
}
@@ -210,7 +211,7 @@ where
let segment_id = worker.segment_id();
on_cancel_sender(worker.engine_mut().cancel_sender());
match worker.run_text(&input).await {
let outcome = match worker.run_text(&input).await {
Ok(lifecycle @ WorkerRunResult::Finished)
| Ok(lifecycle @ WorkerRunResult::Paused)
| Ok(lifecycle @ WorkerRunResult::RolledBack) => Ok(InternalWorkerResult {
@@ -239,7 +240,11 @@ where
identity,
history_entries: store.entries_count(session_id, segment_id),
}),
}
};
worker
.stop_feature_runtime("internal Worker terminal outcome")
.await;
outcome
}
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
@@ -544,7 +549,6 @@ pub(crate) async fn spawn_internal_worker_session(
authority,
} = spec;
manifest.worker.name = format!("internal-{}-{}", identity.kind, identity.run_id);
manifest.memory = None;
let last_usage = Arc::new(Mutex::new(None::<UsageEvent>));
let usage_slot = last_usage.clone();
@@ -575,6 +579,7 @@ pub(crate) async fn spawn_internal_worker_session(
.map_err(|source| InternalWorkerSessionError::Build {
message: source.to_string(),
})?;
worker.disable_manifest_lifecycle_features();
if let Some(session) = inherited_workdir_session {
worker.bind_workdir_session(Some(session));
}
@@ -645,7 +650,6 @@ pub(crate) fn prepare_internal_worker_from_spec(
manifest.mcp = Default::default();
manifest.skills = None;
manifest.compaction = None;
manifest.memory = None;
let mut engine =
Engine::<_, agen::state::Mutable, crate::SessionHistoryMetadata>::new_annotated(client)
@@ -669,6 +673,7 @@ pub(crate) fn prepare_internal_worker_from_spec(
.map_err(|source| InternalWorkerSessionError::Build {
message: source.to_string(),
})?;
worker.disable_manifest_lifecycle_features();
if let Some(session) = inherited_workdir_session {
worker.bind_workdir_session(Some(session));
}
@@ -782,7 +787,8 @@ pub(crate) async fn prepare_internal_worker_session(
};
tokio::spawn(async move {
while let Some(command) = command_rx.recv().await {
let mut stop_done = None;
'actor: while let Some(command) = command_rx.recv().await {
match command {
InternalWorkerSessionCommand::Run(input) => {
actor_in_flight.clear();
@@ -825,21 +831,16 @@ pub(crate) async fn prepare_internal_worker_session(
Some(InternalWorkerSessionCommand::Stop(done)) => {
let _ = cancel_sender.send(()).await;
let _ = (&mut run).await;
actor_in_flight.clear();
status.store(InternalWorkerSessionStatus::Stopped.encode(), std::sync::atomic::Ordering::Release);
let _ = event_tx.send(Event::Status { status: WorkerStatus::Stopped });
let _ = event_tx.send(Event::Shutdown);
state_changed.notify_waiters();
let _ = done.send(());
return;
stop_done = Some(done);
break 'actor;
}
Some(InternalWorkerSessionCommand::Run(_)) => {
// `send` reserves Running atomically, so a second Run cannot be enqueued.
}
None => {
let _ = cancel_sender.send(()).await;
actor_in_flight.clear();
return;
let _ = (&mut run).await;
break 'actor;
}
}
}
@@ -847,22 +848,27 @@ pub(crate) async fn prepare_internal_worker_session(
}
}
InternalWorkerSessionCommand::Stop(done) => {
actor_in_flight.clear();
status.store(
InternalWorkerSessionStatus::Stopped.encode(),
std::sync::atomic::Ordering::Release,
);
let _ = event_tx.send(Event::Status {
status: WorkerStatus::Stopped,
});
let _ = event_tx.send(Event::Shutdown);
state_changed.notify_waiters();
let _ = done.send(());
return;
stop_done = Some(done);
break;
}
}
}
worker
.stop_feature_runtime("internal Worker session stopped")
.await;
actor_in_flight.clear();
status.store(
InternalWorkerSessionStatus::Stopped.encode(),
std::sync::atomic::Ordering::Release,
);
let _ = event_tx.send(Event::Status {
status: WorkerStatus::Stopped,
});
let _ = event_tx.send(Event::Shutdown);
state_changed.notify_waiters();
if let Some(done) = stop_done {
let _ = done.send(());
}
});
Ok(handle)
+364 -91
View File
@@ -15,7 +15,9 @@ use std::sync::{Arc, Mutex};
use agen::Item;
use agen::UsageRecord;
use agen::interceptor::{
Interceptor, PostToolAction, PreRequestAction, PreToolAction, PromptAction, ToolCallInfo,
AssistantTurnEndContext, Interceptor, InterceptorError, InterceptorErrorCategory,
InterceptorResult, PendingHistoryAppendsContext, PostToolAction, PreLlmRequestContext,
PreRequestAction, PreToolAction, PromptAction, PromptSubmitContext, ToolCallInfo,
ToolResultInfo, TurnEndAction,
};
use agen::tool::ToolOutput;
@@ -28,9 +30,9 @@ use crate::compact::usage_tracker::UsageTracker;
use session_store::SystemItem;
use crate::hook::{
AbortInfo, HookPostToolAction, HookPreRequestAction, HookPreToolAction, HookPromptAction,
HookEventKind, HookPostToolAction, HookPreRequestAction, HookPreToolAction, HookPromptAction,
HookRegistry, HookTurnEndAction, PreRequestContext, PreRequestInfo, PromptSubmitInfo,
SystemItemAppendHandle, ToolCallSummary, ToolResultSummary, TurnEndInfo,
RegisteredHook, SystemItemAppendHandle, ToolCallSummary, ToolResultSummary, TurnEndInfo,
};
use crate::ipc::notify_buffer::{NotifyBuffer, build_system_item_with_provenance};
use crate::prompt::catalog::PromptCatalog;
@@ -41,6 +43,32 @@ use agen::token_counter::total_tokens;
/// Maximum number of bytes copied into `TurnEndInfo::final_text_preview`.
const FINAL_TEXT_PREVIEW_LIMIT: usize = 512;
const INLINE_HOOK_CHAIN_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(30);
async fn call_hook_before_deadline<E: HookEventKind>(
hook: &RegisteredHook<E>,
input: &E::Input,
deadline: tokio::time::Instant,
) -> Result<Option<E::Output>, InterceptorError> {
let remaining = deadline.saturating_duration_since(tokio::time::Instant::now());
if remaining.is_zero() {
return Err(InterceptorError::new(
InterceptorErrorCategory::Policy,
"Worker hook chain exceeded its host-enforced lifecycle deadline",
));
}
tokio::time::timeout(remaining, hook.call_optional(input))
.await
.map_err(|_| {
InterceptorError::new(
InterceptorErrorCategory::Policy,
"Worker hook chain exceeded its host-enforced lifecycle deadline",
)
})?
.map_err(|error| {
InterceptorError::new(InterceptorErrorCategory::Dependency, error.to_string())
})
}
pub(crate) struct WorkerInterceptor {
registry: Arc<HookRegistry>,
@@ -231,8 +259,12 @@ impl WorkerInterceptor {
}
#[async_trait]
impl Interceptor for WorkerInterceptor {
async fn on_prompt_submit(&self, item: &mut Item) -> PromptAction {
impl Interceptor<SessionHistoryMetadata> for WorkerInterceptor {
async fn on_prompt_submit(
&self,
context: PromptSubmitContext<'_, SessionHistoryMetadata>,
) -> InterceptorResult<PromptAction> {
let item = context.item;
let turn_index = self.next_turn_index.fetch_add(1, Ordering::Relaxed);
self.tool_calls_this_turn.store(0, Ordering::Relaxed);
@@ -240,19 +272,27 @@ impl Interceptor for WorkerInterceptor {
input_text: extract_message_text(item).unwrap_or_default(),
turn_index,
};
let deadline = tokio::time::Instant::now() + INLINE_HOOK_CHAIN_TIMEOUT;
let mut cancellations = Vec::new();
for hook in &self.registry.on_prompt_submit {
let action = hook.call(&info).await;
if !matches!(action, HookPromptAction::Continue) {
return action.into();
let Some(action) = call_hook_before_deadline(hook, &info, deadline).await? else {
continue;
};
if let HookPromptAction::Cancel(reason) = action {
cancellations.push(reason);
}
}
cancellations.sort();
if let Some(reason) = cancellations.into_iter().next() {
return Ok(PromptAction::Cancel(reason));
}
let mut extras: Vec<SystemItem> = std::mem::take(
&mut *self
.pending_attachments
.lock()
.expect("pending_attachments poisoned"),
);
if extras.is_empty() {
Ok(if extras.is_empty() {
PromptAction::Continue
} else {
// Commit the typed system items first, then hand the
@@ -266,10 +306,13 @@ impl Interceptor for WorkerInterceptor {
Ok(()) => PromptAction::ContinueWith(items),
Err(error) => PromptAction::Cancel(format!("session persistence failed: {error}")),
}
}
})
}
async fn pending_history_appends(&self) -> Result<Vec<Item>, String> {
async fn pending_history_appends(
&self,
_context: PendingHistoryAppendsContext<'_, SessionHistoryMetadata>,
) -> InterceptorResult<Vec<Item>> {
let drained = self.pending_notifies.drain();
if drained.is_empty() {
return Ok(Vec::new());
@@ -295,7 +338,10 @@ impl Interceptor for WorkerInterceptor {
Ok(system_item) => system_item,
Err(error) => {
self.pending_notifies.requeue_front(drained);
return Err(format!("failed to render notify_wrapper: {error}"));
return Err(InterceptorError::new(
InterceptorErrorCategory::Dependency,
format!("failed to render notify_wrapper: {error}"),
));
}
};
items.push(system_item.to_history_item());
@@ -303,15 +349,22 @@ impl Interceptor for WorkerInterceptor {
}
if let Err(error) = self.commit_system_items(&system_items) {
self.pending_notifies.requeue_front(drained);
return Err(format!("session persistence failed: {error}"));
return Err(InterceptorError::new(
InterceptorErrorCategory::Dependency,
format!("session persistence failed: {error}"),
));
}
Ok(items)
}
async fn pre_llm_request(&self, context: &mut Vec<Item>) -> PreRequestAction {
async fn pre_llm_request(
&self,
context: PreLlmRequestContext<'_, SessionHistoryMetadata>,
) -> InterceptorResult<PreRequestAction> {
let context = context.items;
let initial_tokens = self.estimated_tokens(context);
if self.request_threshold_exceeded(initial_tokens, context) {
return PreRequestAction::Yield;
return Ok(PreRequestAction::Yield);
}
let info = PreRequestInfo {
item_count: context.len(),
@@ -325,12 +378,28 @@ impl Interceptor for WorkerInterceptor {
.as_ref()
.map(|_| SystemItemAppendHandle::new(Arc::clone(&pending_hook_system_items)));
let hook_context = PreRequestContext::new(info, system_item_sink);
let deadline = tokio::time::Instant::now() + INLINE_HOOK_CHAIN_TIMEOUT;
let mut cancellations = Vec::new();
let mut should_yield = false;
for hook in &self.registry.pre_llm_request {
let action = hook.call(&hook_context).await;
if !matches!(action, HookPreRequestAction::Continue) {
return action.into();
let Some(action) = call_hook_before_deadline(hook, &hook_context, deadline).await?
else {
continue;
};
match action {
HookPreRequestAction::Continue => {}
HookPreRequestAction::Yield => should_yield = true,
HookPreRequestAction::Cancel(reason) => cancellations.push(reason),
}
}
cancellations.sort();
if let Some(reason) = cancellations.into_iter().next() {
return Ok(PreRequestAction::Cancel(reason));
}
if should_yield {
return Ok(PreRequestAction::Yield);
}
let mut system_items: Vec<SystemItem> = std::mem::take(
&mut *pending_hook_system_items
@@ -353,44 +422,73 @@ impl Interceptor for WorkerInterceptor {
if self.request_threshold_exceeded(current_tokens, effective_context.as_ref()) {
if let Err(error) = self.commit_system_items(&system_items) {
return PreRequestAction::Cancel(format!("session persistence failed: {error}"));
return Ok(PreRequestAction::Cancel(format!(
"session persistence failed: {error}"
)));
}
return if appended_items.is_empty() {
return Ok(if appended_items.is_empty() {
PreRequestAction::Yield
} else {
PreRequestAction::YieldWith(appended_items)
};
});
}
if let Some(usage_tracker) = self.usage_tracker.as_ref() {
usage_tracker.note_request(effective_context.len());
}
if system_items.is_empty() {
return PreRequestAction::Continue;
return Ok(PreRequestAction::Continue);
}
match self.commit_system_items(&system_items) {
Ok(match self.commit_system_items(&system_items) {
Ok(()) => PreRequestAction::ContinueWith(appended_items),
Err(error) => PreRequestAction::Cancel(format!("session persistence failed: {error}")),
}
})
}
async fn pre_tool_call(&self, info: &mut ToolCallInfo) -> PreToolAction {
async fn pre_tool_call(
&self,
info: &mut ToolCallInfo<'_, SessionHistoryMetadata>,
) -> InterceptorResult<PreToolAction> {
let summary = ToolCallSummary {
call_id: info.call.id.clone(),
tool_name: info.call.name.clone(),
arguments: info.call.input.clone(),
};
let deadline = tokio::time::Instant::now() + INLINE_HOOK_CHAIN_TIMEOUT;
let mut aborts = Vec::new();
let mut should_pause = false;
let mut denials = Vec::new();
for hook in &self.registry.pre_tool_call {
let action = hook.call(&summary).await;
if !matches!(action, HookPreToolAction::Continue) {
return action.into_worker_action(summary.call_id.clone());
let Some(action) = call_hook_before_deadline(hook, &summary, deadline).await? else {
continue;
};
match action {
HookPreToolAction::Continue => {}
HookPreToolAction::Pause => should_pause = true,
HookPreToolAction::Deny(reason) => denials.push(reason),
HookPreToolAction::Abort(reason) => aborts.push(reason),
}
}
aborts.sort();
if let Some(reason) = aborts.into_iter().next() {
return Ok(HookPreToolAction::Abort(reason).into_worker_action(summary.call_id.clone()));
}
if should_pause {
return Ok(PreToolAction::Pause);
}
denials.sort();
if let Some(reason) = denials.into_iter().next() {
return Ok(HookPreToolAction::Deny(reason).into_worker_action(summary.call_id.clone()));
}
self.tool_calls_this_turn.fetch_add(1, Ordering::Relaxed);
PreToolAction::Continue
Ok(PreToolAction::Continue)
}
async fn post_tool_call(&self, info: &mut ToolResultInfo) -> PostToolAction {
async fn post_tool_call(
&self,
info: &ToolResultInfo<'_, SessionHistoryMetadata>,
) -> InterceptorResult<PostToolAction> {
let summary = ToolResultSummary {
call_id: info.result.tool_use_id.clone(),
tool_name: info.call.name.clone(),
@@ -402,21 +500,34 @@ impl Interceptor for WorkerInterceptor {
attachments: Vec::new(),
},
};
let deadline = tokio::time::Instant::now() + INLINE_HOOK_CHAIN_TIMEOUT;
let mut aborts = Vec::new();
for hook in &self.registry.post_tool_call {
let action = hook.call(&summary).await;
if !matches!(action, HookPostToolAction::Continue) {
return action.into();
let Some(action) = call_hook_before_deadline(hook, &summary, deadline).await? else {
continue;
};
if let HookPostToolAction::Abort(reason) = action {
aborts.push(reason);
}
}
PostToolAction::Continue
aborts.sort();
if let Some(reason) = aborts.into_iter().next() {
return Ok(PostToolAction::Abort(reason));
}
Ok(PostToolAction::Continue)
}
async fn on_turn_end(&self, history: &[Item]) -> TurnEndAction {
async fn on_assistant_turn_end(
&self,
context: AssistantTurnEndContext<'_, SessionHistoryMetadata>,
) -> InterceptorResult<TurnEndAction> {
let history = context.history;
let final_text_preview = history
.iter()
.rev()
.find(|i| i.is_assistant_message())
.and_then(extract_message_text)
.find(|entry| entry.item.is_assistant_message())
.and_then(|entry| extract_message_text(&entry.item))
.map(|t| preview(&t, FINAL_TEXT_PREVIEW_LIMIT))
.unwrap_or_default();
let info = TurnEndInfo {
@@ -424,22 +535,20 @@ impl Interceptor for WorkerInterceptor {
tool_calls_count: self.tool_calls_this_turn.load(Ordering::Relaxed),
final_text_preview,
};
let deadline = tokio::time::Instant::now() + INLINE_HOOK_CHAIN_TIMEOUT;
let mut should_pause = false;
for hook in &self.registry.on_turn_end {
let action = hook.call(&info).await;
if !matches!(action, HookTurnEndAction::Finish) {
return action.into();
let Some(action) = call_hook_before_deadline(hook, &info, deadline).await? else {
continue;
};
if matches!(action, HookTurnEndAction::Pause) {
should_pause = true;
}
}
TurnEndAction::Finish
}
async fn on_abort(&self, reason: &str) {
let info = AbortInfo {
reason: reason.to_string(),
};
for hook in &self.registry.on_abort {
hook.call(&info).await;
if should_pause {
return Ok(TurnEndAction::Pause);
}
Ok(TurnEndAction::Finish)
}
}
@@ -509,6 +618,7 @@ mod tests {
Hook, HookPostToolAction, HookPreRequestAction, HookPreToolAction, HookRegistryBuilder,
HookTurnEndAction, OnTurnEnd, PostToolCall, PreLlmRequest, PreToolCall,
};
use crate::session_history::{WorkerHistoryProvenance, history_entry};
fn test_prompts() -> Arc<ArcSwap<PromptCatalog>> {
Arc::new(ArcSwap::from(PromptCatalog::builtins_only().unwrap()))
@@ -518,9 +628,12 @@ mod tests {
#[async_trait]
impl Hook<PreLlmRequest> for CountingHook {
async fn call(&self, _info: &PreRequestContext) -> HookPreRequestAction {
async fn call(
&self,
_info: &PreRequestContext,
) -> Result<HookPreRequestAction, crate::hook::HookError> {
self.0.fetch_add(1, Ordering::Relaxed);
HookPreRequestAction::Continue
Ok(HookPreRequestAction::Continue)
}
}
@@ -559,16 +672,22 @@ mod tests {
#[async_trait]
impl Hook<PreLlmRequest> for AppendingPreRequestHook {
async fn call(&self, input: &PreRequestContext) -> HookPreRequestAction {
async fn call(
&self,
input: &PreRequestContext,
) -> Result<HookPreRequestAction, crate::hook::HookError> {
if let Some(system_items) = input.system_items() {
self.saw_handle.store(true, Ordering::Relaxed);
system_items.append_task_reminder("hook reminder");
}
HookPreRequestAction::Continue
Ok(HookPreRequestAction::Continue)
}
}
fn task_tool_call_info(name: &str, input: serde_json::Value) -> ToolCallInfo {
fn task_tool_call_info(
name: &str,
input: serde_json::Value,
) -> ToolCallInfo<'static, SessionHistoryMetadata> {
let def = crate::feature::builtin::task::task_tools(
crate::feature::builtin::task::TaskStore::new(),
)
@@ -580,6 +699,8 @@ mod tests {
.expect("task tool definition");
let (meta, tool) = def();
ToolCallInfo {
invocation: Default::default(),
history: &[],
call: agen::tool::ToolCall {
id: "call-id".into(),
name: name.into(),
@@ -623,7 +744,14 @@ mod tests {
None,
);
let mut ctx = ctx_items;
let action = interceptor.pre_llm_request(&mut ctx).await;
let action = interceptor
.pre_llm_request(PreLlmRequestContext {
invocation: Default::default(),
items: &mut ctx,
history: &[],
})
.await
.unwrap();
assert!(matches!(action, PreRequestAction::Yield));
// Hook must not run when an internal mechanism short-circuits first.
@@ -655,7 +783,14 @@ mod tests {
})),
);
let mut ctx = ctx_items;
let action = interceptor.pre_llm_request(&mut ctx).await;
let action = interceptor
.pre_llm_request(PreLlmRequestContext {
invocation: Default::default(),
items: &mut ctx,
history: &[],
})
.await
.unwrap();
match action {
PreRequestAction::YieldWith(items) => assert_eq!(items.len(), 1),
@@ -692,7 +827,14 @@ mod tests {
)
.with_usage_tracker(usage_tracker);
let mut ctx = ctx_items;
let action = interceptor.pre_llm_request(&mut ctx).await;
let action = interceptor
.pre_llm_request(PreLlmRequestContext {
invocation: Default::default(),
items: &mut ctx,
history: &[],
})
.await
.unwrap();
assert!(matches!(action, PreRequestAction::Yield));
}
@@ -716,7 +858,14 @@ mod tests {
None,
);
let mut ctx = ctx_items;
let action = interceptor.pre_llm_request(&mut ctx).await;
let action = interceptor
.pre_llm_request(PreLlmRequestContext {
invocation: Default::default(),
items: &mut ctx,
history: &[],
})
.await
.unwrap();
assert!(matches!(action, PreRequestAction::Continue));
assert_eq!(count.load(Ordering::Relaxed), 1);
@@ -757,7 +906,14 @@ mod tests {
None,
);
let mut ctx = ctx_items;
let action = interceptor.pre_llm_request(&mut ctx).await;
let action = interceptor
.pre_llm_request(PreLlmRequestContext {
invocation: Default::default(),
items: &mut ctx,
history: &[],
})
.await
.unwrap();
assert!(matches!(action, PreRequestAction::Continue));
assert_eq!(count.load(Ordering::Relaxed), 1);
@@ -784,7 +940,14 @@ mod tests {
None,
);
let mut ctx = ctx_items;
let action = interceptor.pre_llm_request(&mut ctx).await;
let action = interceptor
.pre_llm_request(PreLlmRequestContext {
invocation: Default::default(),
items: &mut ctx,
history: &[],
})
.await
.unwrap();
assert!(matches!(action, PreRequestAction::Continue));
assert_eq!(count.load(Ordering::Relaxed), 1);
@@ -805,7 +968,14 @@ mod tests {
None,
);
let mut ctx: Vec<Item> = Vec::new();
let action = interceptor.pre_llm_request(&mut ctx).await;
let action = interceptor
.pre_llm_request(PreLlmRequestContext {
invocation: Default::default(),
items: &mut ctx,
history: &[],
})
.await
.unwrap();
assert!(matches!(action, PreRequestAction::Continue));
assert_eq!(count.load(Ordering::Relaxed), 1);
@@ -834,7 +1004,14 @@ mod tests {
);
let mut ctx: Vec<Item> = Vec::new();
let action = interceptor.pre_llm_request(&mut ctx).await;
let action = interceptor
.pre_llm_request(PreLlmRequestContext {
invocation: Default::default(),
items: &mut ctx,
history: &[],
})
.await
.unwrap();
assert!(saw_handle.load(Ordering::Relaxed));
let PreRequestAction::ContinueWith(items) = action else {
@@ -881,7 +1058,14 @@ mod tests {
);
let mut ctx: Vec<Item> = Vec::new();
let action = interceptor.pre_llm_request(&mut ctx).await;
let action = interceptor
.pre_llm_request(PreLlmRequestContext {
invocation: Default::default(),
items: &mut ctx,
history: &[],
})
.await
.unwrap();
assert!(!saw_handle.load(Ordering::Relaxed));
assert!(matches!(action, PreRequestAction::Continue));
@@ -891,33 +1075,42 @@ mod tests {
#[async_trait]
impl Hook<PreLlmRequest> for AbortingHook {
async fn call(&self, _info: &PreRequestContext) -> HookPreRequestAction {
async fn call(
&self,
_info: &PreRequestContext,
) -> Result<HookPreRequestAction, crate::hook::HookError> {
self.0.store(true, Ordering::Relaxed);
HookPreRequestAction::Cancel("nope".into())
Ok(HookPreRequestAction::Cancel("nope".into()))
}
}
#[tokio::test]
async fn public_pre_tool_hook_deny_becomes_synthetic_error_and_short_circuits() {
async fn public_pre_tool_hook_denials_compose_without_short_circuiting() {
struct DenyToolHook(Arc<AtomicUsize>);
struct CountingToolHook(Arc<AtomicUsize>);
#[async_trait]
impl Hook<PreToolCall> for DenyToolHook {
async fn call(&self, input: &ToolCallSummary) -> HookPreToolAction {
async fn call(
&self,
input: &ToolCallSummary,
) -> Result<HookPreToolAction, crate::hook::HookError> {
self.0.fetch_add(1, Ordering::Relaxed);
assert_eq!(input.call_id, "call-id");
assert_eq!(input.tool_name, "TaskList");
assert_eq!(input.arguments, serde_json::json!({"scope": "all"}));
HookPreToolAction::Deny("blocked by public hook".into())
Ok(HookPreToolAction::Deny("blocked by public hook".into()))
}
}
#[async_trait]
impl Hook<PreToolCall> for CountingToolHook {
async fn call(&self, _input: &ToolCallSummary) -> HookPreToolAction {
async fn call(
&self,
_input: &ToolCallSummary,
) -> Result<HookPreToolAction, crate::hook::HookError> {
self.0.fetch_add(1, Ordering::Relaxed);
HookPreToolAction::Continue
Ok(HookPreToolAction::Continue)
}
}
@@ -938,7 +1131,7 @@ mod tests {
);
let mut info = task_tool_call_info("TaskList", serde_json::json!({"scope": "all"}));
let action = interceptor.pre_tool_call(&mut info).await;
let action = interceptor.pre_tool_call(&mut info).await.unwrap();
match action {
PreToolAction::SyntheticResult(result) => {
@@ -950,7 +1143,7 @@ mod tests {
other => panic!("expected synthetic denial, got {other:?}"),
}
assert_eq!(first_count.load(Ordering::Relaxed), 1);
assert_eq!(second_count.load(Ordering::Relaxed), 0);
assert_eq!(second_count.load(Ordering::Relaxed), 1);
}
#[tokio::test]
@@ -959,14 +1152,17 @@ mod tests {
#[async_trait]
impl Hook<PostToolCall> for AbortAfterToolHook {
async fn call(&self, input: &ToolResultSummary) -> HookPostToolAction {
async fn call(
&self,
input: &ToolResultSummary,
) -> Result<HookPostToolAction, crate::hook::HookError> {
self.0.fetch_add(1, Ordering::Relaxed);
assert_eq!(input.call_id, "call-id");
assert_eq!(input.tool_name, "TaskList");
assert!(!input.is_error);
assert_eq!(input.output.summary, "ok");
assert_eq!(input.output.content.as_deref(), Some("full"));
HookPostToolAction::Abort("post tool abort".into())
Ok(HookPostToolAction::Abort("post tool abort".into()))
}
}
@@ -984,7 +1180,9 @@ mod tests {
None,
);
let info = task_tool_call_info("TaskList", serde_json::json!({}));
let mut result_info = ToolResultInfo {
let result_info = ToolResultInfo {
invocation: Default::default(),
history: &[],
call: info.call,
result: agen::tool::ToolResult::from_output(
"call-id",
@@ -1000,7 +1198,7 @@ mod tests {
context: info.context,
};
let action = interceptor.post_tool_call(&mut result_info).await;
let action = interceptor.post_tool_call(&result_info).await.unwrap();
assert_eq!(action, PostToolAction::Abort("post tool abort".to_string()));
assert_eq!(count.load(Ordering::Relaxed), 1);
@@ -1012,12 +1210,15 @@ mod tests {
#[async_trait]
impl Hook<OnTurnEnd> for PauseTurnEndHook {
async fn call(&self, input: &TurnEndInfo) -> HookTurnEndAction {
async fn call(
&self,
input: &TurnEndInfo,
) -> Result<HookTurnEndAction, crate::hook::HookError> {
self.0.fetch_add(1, Ordering::Relaxed);
assert_eq!(input.turn_index, 0);
assert_eq!(input.tool_calls_count, 0);
assert_eq!(input.final_text_preview, "done");
HookTurnEndAction::Pause
Ok(HookTurnEndAction::Pause)
}
}
@@ -1034,9 +1235,25 @@ mod tests {
test_prompts(),
None,
);
let history = vec![Item::user_message("hi"), Item::assistant_message("done")];
let action = interceptor.on_turn_end(&history).await;
let history = vec![
history_entry(
Item::user_message("hi"),
WorkerHistoryProvenance::LegacyUnknown,
),
history_entry(
Item::assistant_message("done"),
WorkerHistoryProvenance::LegacyUnknown,
),
];
let action = interceptor
.on_assistant_turn_end(AssistantTurnEndContext {
invocation: Default::default(),
assistant_entries: &history[1..],
history: &history,
tool_calls: &[],
})
.await
.unwrap();
assert!(matches!(action, TurnEndAction::Pause));
assert_eq!(count.load(Ordering::Relaxed), 1);
@@ -1073,7 +1290,14 @@ mod tests {
let ctx_items = vec![Item::user_message("hi")];
for _ in 0..23 {
let mut ctx = ctx_items.clone();
let action = interceptor.pre_llm_request(&mut ctx).await;
let action = interceptor
.pre_llm_request(PreLlmRequestContext {
invocation: Default::default(),
items: &mut ctx,
history: &[],
})
.await
.unwrap();
assert!(matches!(action, PreRequestAction::Continue));
usage_tracker.record_usage(&agen::event::UsageEvent {
input_tokens: Some(10),
@@ -1085,7 +1309,14 @@ mod tests {
}
let mut ctx = ctx_items.clone();
let action = interceptor.pre_llm_request(&mut ctx).await;
let action = interceptor
.pre_llm_request(PreLlmRequestContext {
invocation: Default::default(),
items: &mut ctx,
history: &[],
})
.await
.unwrap();
let appended_len = match action {
PreRequestAction::ContinueWith(items) => items.len(),
other => panic!("expected reminder append, got {other:?}"),
@@ -1158,7 +1389,13 @@ mod tests {
));
buffer.push_notify("updated".to_string(), false);
let appends = interceptor.pending_history_appends().await.unwrap();
let appends = interceptor
.pending_history_appends(PendingHistoryAppendsContext {
invocation: Default::default(),
history: &[],
})
.await
.unwrap();
assert_eq!(appends.len(), 1);
assert!(format!("{:?}", appends[0]).contains("CURRENT-PROJECTION updated"));
let committed = committed.lock().unwrap();
@@ -1208,9 +1445,19 @@ mod tests {
));
buffer.push_notify("must persist".to_string(), false);
let error = interceptor.pending_history_appends().await.unwrap_err();
let error = interceptor
.pending_history_appends(PendingHistoryAppendsContext {
invocation: Default::default(),
history: &[],
})
.await
.unwrap_err();
assert!(error.contains("failed to render notify_wrapper"));
assert!(
error
.diagnostic()
.contains("failed to render notify_wrapper")
);
let requeued = buffer.drain();
assert_eq!(requeued.len(), 1);
}
@@ -1232,7 +1479,13 @@ mod tests {
None,
);
let items = interceptor.pending_history_appends().await.unwrap();
let items = interceptor
.pending_history_appends(PendingHistoryAppendsContext {
invocation: Default::default(),
history: &[],
})
.await
.unwrap();
assert_eq!(items.len(), 2);
let first = items[0].as_text().unwrap_or_default();
let second = items[1].as_text().unwrap_or_default();
@@ -1246,7 +1499,13 @@ mod tests {
);
// Empty buffer → empty Vec (no synthesised items).
let again = interceptor.pending_history_appends().await.unwrap();
let again = interceptor
.pending_history_appends(PendingHistoryAppendsContext {
invocation: Default::default(),
history: &[],
})
.await
.unwrap();
assert!(again.is_empty());
}
@@ -1269,7 +1528,14 @@ mod tests {
None,
);
let mut ctx: Vec<Item> = vec![Item::user_message("hi")];
let action = interceptor.pre_llm_request(&mut ctx).await;
let action = interceptor
.pre_llm_request(PreLlmRequestContext {
invocation: Default::default(),
items: &mut ctx,
history: &[],
})
.await
.unwrap();
assert!(matches!(action, PreRequestAction::Continue));
assert_eq!(ctx.len(), 1, "pre_llm_request must not append notifies");
@@ -1299,10 +1565,17 @@ mod tests {
None,
);
let mut ctx: Vec<Item> = Vec::new();
let action = interceptor.pre_llm_request(&mut ctx).await;
let action = interceptor
.pre_llm_request(PreLlmRequestContext {
invocation: Default::default(),
items: &mut ctx,
history: &[],
})
.await
.unwrap();
assert!(matches!(action, PreRequestAction::Cancel(_)));
assert!(first_called.load(Ordering::Relaxed));
assert_eq!(second_count.load(Ordering::Relaxed), 0);
assert_eq!(second_count.load(Ordering::Relaxed), 1);
}
}
+1
View File
@@ -50,6 +50,7 @@ pub use prompt::system::{SystemPromptContext, SystemPromptError, SystemPromptTem
pub use protocol::{ErrorCode, Event, Method, TurnResult, WorkerStatus};
pub use runtime::dir::RuntimeDir;
pub use segment_log_sink::SegmentLogSink;
pub use session_capture::SessionEntryRef;
pub use session_history::{
SessionHistoryDerivation, SessionHistoryEntryId, SessionHistoryMetadata,
WorkerHistoryProvenance, WorkerSubjectSnapshot,
+8 -5
View File
@@ -45,14 +45,17 @@ impl<C: LlmClient + 'static, St: Store> Worker<C, St> {
#[async_trait]
impl Hook<PreToolCall> for PermissionHook {
async fn call(&self, input: &ToolCallSummary) -> HookPreToolAction {
match self.action_for(input) {
async fn call(
&self,
input: &ToolCallSummary,
) -> Result<HookPreToolAction, crate::hook::HookError> {
Ok(match self.action_for(input) {
ToolPermissionAction::Allow => HookPreToolAction::Continue,
ToolPermissionAction::Deny => HookPreToolAction::Deny(permission_denied_message(input)),
ToolPermissionAction::Ask => {
HookPreToolAction::Deny(permission_ask_unsupported_message(input))
}
}
})
}
}
@@ -174,7 +177,7 @@ mod tests {
))
.await;
match denied {
HookPreToolAction::Deny(message) => {
Ok(HookPreToolAction::Deny(message)) => {
assert!(message.contains("permission denied"));
assert!(message.contains("Bash"));
}
@@ -192,7 +195,7 @@ mod tests {
))
.await;
match asked {
HookPreToolAction::Deny(message) => {
Ok(HookPreToolAction::Deny(message)) => {
assert!(message.contains("permission ask unsupported"));
assert!(message.contains("denied fail-closed"));
}
+4 -4
View File
@@ -23,14 +23,14 @@ const OVERVIEW_ANCHOR_STRIDE: usize = 8;
#[derive(Debug, Clone, PartialEq, Eq, Hash, Serialize, Deserialize)]
#[serde(transparent)]
pub(crate) struct SessionEntryRef(String);
pub struct SessionEntryRef(String);
impl SessionEntryRef {
pub(crate) fn from_history_entry_id(entry_id: &crate::SessionHistoryEntryId) -> Self {
pub fn from_history_entry_id(entry_id: &crate::SessionHistoryEntryId) -> Self {
Self(format!("E{}", entry_id.0))
}
pub(crate) fn parse(value: &str) -> Option<Self> {
pub fn parse(value: &str) -> Option<Self> {
let suffix = value.strip_prefix('E')?;
if suffix.is_empty()
|| suffix.len() > 64
@@ -43,7 +43,7 @@ impl SessionEntryRef {
Some(Self(value.to_string()))
}
pub(crate) fn as_str(&self) -> &str {
pub fn as_str(&self) -> &str {
&self.0
}
+5 -2
View File
@@ -70,9 +70,12 @@ impl TicketIntakeReadyShutdownHook {
#[async_trait]
impl Hook<PostToolCall> for TicketIntakeReadyShutdownHook {
async fn call(&self, info: &ToolResultSummary) -> HookPostToolAction {
async fn call(
&self,
info: &ToolResultSummary,
) -> Result<HookPostToolAction, crate::hook::HookError> {
self.observe_tool_result(info);
HookPostToolAction::Continue
Ok(HookPostToolAction::Continue)
}
}
+28 -125
View File
@@ -2,127 +2,11 @@ use serde::{Deserialize, Serialize};
use crate::worker::WorkspaceClient;
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
#[serde(rename_all = "snake_case")]
pub enum SkillDiagnosticSeverity {
Error,
Warning,
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
pub struct SkillDiagnostic {
pub severity: SkillDiagnosticSeverity,
pub code: String,
pub message: String,
/// Path-free authority/provenance label such as `builtin:foo` or `workspace:foo`.
#[serde(skip_serializing_if = "Option::is_none")]
pub source: Option<String>,
}
impl SkillDiagnostic {
pub fn error(
code: impl Into<String>,
message: impl Into<String>,
source: Option<String>,
) -> Self {
Self {
severity: SkillDiagnosticSeverity::Error,
code: code.into(),
message: message.into(),
source,
}
}
pub fn warning(
code: impl Into<String>,
message: impl Into<String>,
source: Option<String>,
) -> Self {
Self {
severity: SkillDiagnosticSeverity::Warning,
code: code.into(),
message: message.into(),
source,
}
}
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
#[serde(rename_all = "snake_case")]
pub enum SkillSourceKind {
Builtin,
Workspace,
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
pub struct SkillProvenance {
pub kind: SkillSourceKind,
/// Stable id: `builtin:<name>` or `workspace:<name>`.
pub id: String,
/// Virtual config/resource path. Never an absolute host filesystem path.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub virtual_path: Option<String>,
/// Active Workspace config revision for Workspace Skills.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub revision: Option<u64>,
/// Digest of the immutable `SKILL.md` source.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub source_digest: Option<String>,
/// Digest of the active virtual config tree snapshot.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub tree_digest: Option<String>,
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
pub struct SkillResourceRef {
pub kind: String,
/// Skill-relative resource name/path. Never an absolute filesystem path.
pub name: String,
pub supported: bool,
#[serde(skip_serializing_if = "Option::is_none")]
pub diagnostic: Option<String>,
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
pub struct SkillCatalogEntry {
pub name: String,
pub description: String,
pub provenance: SkillProvenance,
#[serde(default)]
pub overrides: Vec<SkillProvenance>,
#[serde(default)]
pub diagnostics: Vec<SkillDiagnostic>,
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
pub struct SkillCatalogResponse {
/// Authority label for diagnostics; callers must not interpret it as a path.
pub authority: String,
#[serde(default)]
pub entries: Vec<SkillCatalogEntry>,
#[serde(default)]
pub diagnostics: Vec<SkillDiagnostic>,
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
pub struct SkillDetailResponse {
pub name: String,
pub description: String,
pub provenance: SkillProvenance,
#[serde(default)]
pub overrides: Vec<SkillProvenance>,
#[serde(default)]
pub diagnostics: Vec<SkillDiagnostic>,
/// Imported Markdown content with YAML frontmatter delimiters removed.
/// This is intentionally omitted from catalog responses.
pub body: String,
#[serde(default)]
pub allowed_tools: Vec<String>,
/// Explicitly documents that allowed-tools is parsed only as an experimental hint.
pub allowed_tools_status: String,
#[serde(default)]
pub resources: Vec<SkillResourceRef>,
}
pub use workspace_api::{
SkillActivationStatus, SkillCatalogEntry, SkillCatalogResponse, SkillDetailResponse,
SkillDiagnostic, SkillDiagnosticSeverity, SkillProjectionIdentity, SkillProjectionStatus,
SkillProvenance, SkillResourceRef, SkillSourceKind,
};
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
pub struct SkillActivationResponse {
@@ -144,6 +28,8 @@ pub enum SkillClientError {
Request(#[from] crate::worker::WorkspaceClientError),
#[error("Skill API response JSON is invalid: {0}")]
Json(#[from] serde_json::Error),
#[error("Skill API response violates the shared contract: {0}")]
InvalidResponse(#[from] workspace_api::SkillApiValidationError),
#[error("Skill API returned HTTP {status}: {body}")]
Http {
status: reqwest::StatusCode,
@@ -155,11 +41,15 @@ pub enum SkillClientError {
impl dyn WorkspaceClient + '_ {
pub fn list_skills(&self) -> Result<SkillCatalogResponse, SkillClientError> {
self.get_skill_json("skills")
let response: SkillCatalogResponse = self.get_skill_json("skills")?;
response.validate()?;
Ok(response)
}
pub fn read_skill(&self, name: &str) -> Result<SkillDetailResponse, SkillClientError> {
self.get_skill_json(&format!("skills/{name}"))
let response: SkillDetailResponse = self.get_skill_json(&format!("skills/{name}"))?;
response.validate()?;
Ok(response)
}
pub fn activate_skill(&self, name: &str) -> Result<SkillActivationResponse, SkillClientError> {
@@ -229,11 +119,24 @@ mod tests {
assert_eq!(worker_header, None);
assert_eq!(authorization, None);
let body = serde_json::json!({
"authority": "workspace-backend-skills-v0",
"authority": "workspace-config-skills-v1",
"projection": {
"config_revision": 7,
"tree_digest": "tree-digest"
},
"entries": [{
"name": "triage-errors",
"description": "Use when triaging errors.",
"provenance": { "kind": "workspace", "id": "workspace:triage-errors" },
"activation_status": "active",
"projection_status": "valid",
"provenance": {
"kind": "workspace",
"id": "workspace:triage-errors",
"virtual_path": "skills/triage-errors/SKILL.md",
"revision": 7,
"source_digest": "source-digest",
"tree_digest": "tree-digest"
},
"overrides": [],
"diagnostics": []
}],
+57 -5
View File
@@ -425,6 +425,8 @@ impl Tool for SubWorkerSpawnTool {
WorkerManifestConfig::resolution_defaults().merge(child_config),
)
.map_err(|error| ToolError::ExecutionFailed(format!("resolve child manifest: {error}")))?;
bind_child_memory_settings(&self.spawner_manifest, &mut child_manifest)
.map_err(ToolError::ExecutionFailed)?;
// Delegated children stay bound to their scoped session and cannot use
// Workspace attachment tools to replace it with parent-level authority.
child_manifest.feature.manage_workdir.enabled = false;
@@ -827,6 +829,33 @@ fn profile_error_with_available(error: ProfileError, available: &AvailableProfil
)
}
fn bind_child_memory_settings(
parent: &manifest::WorkerManifest,
child: &mut manifest::WorkerManifest,
) -> Result<(), String> {
if !child.feature.memory.profile.enabled {
return child
.feature
.memory
.validate_execution()
.map_err(str::to_string);
}
let workspace_settings = parent.feature.memory.workspace_settings().ok_or_else(|| {
"enabled child Memory feature requires the parent's trusted Workspace settings snapshot"
.to_string()
})?;
child
.feature
.memory
.bind_workspace_settings(workspace_settings)
.map_err(str::to_string)?;
child
.feature
.memory
.validate_execution()
.map_err(str::to_string)
}
fn manifest_to_reusable_config(manifest: &WorkerManifest) -> WorkerManifestConfig {
WorkerManifestConfig {
worker: WorkerMetaConfig {
@@ -894,7 +923,6 @@ fn manifest_to_reusable_config(manifest: &WorkerManifest) -> WorkerManifestConfi
model: c.model.clone(),
}),
web: manifest.web.clone(),
memory: manifest.memory.clone(),
skills: manifest.skills.clone(),
}
}
@@ -1091,10 +1119,7 @@ enabled = true
thread = true
[feature.memory]
enabled = true
[memory]
extract_threshold = 4000
enabled = false
"#;
#[tokio::test]
@@ -1526,6 +1551,33 @@ extract_threshold = 4000
.unwrap()
}
#[test]
fn child_memory_inherits_only_the_parents_trusted_settings_snapshot() {
let temp = tempfile::tempdir().unwrap();
let mut parent = parent_manifest(temp.path(), None);
parent.feature.memory.profile.enabled = true;
parent
.feature
.memory
.bind_workspace_settings(manifest::WorkspaceMemorySettingsSnapshot {
workspace_id: "workspace-1".to_string(),
settings_revision: 4,
language: "日本語".to_string(),
})
.unwrap();
let mut child = parent.clone();
child.feature.memory.workspace_settings = None;
bind_child_memory_settings(&parent, &mut child).unwrap();
assert_eq!(
child.feature.memory.workspace_settings(),
parent.feature.memory.workspace_settings()
);
child.feature.memory.profile.enabled = false;
assert!(bind_child_memory_settings(&parent, &mut child).is_err());
}
fn write_project_profile_registry(
project: &Path,
default: Option<&str>,
+514 -1457
View File
File diff suppressed because it is too large Load Diff
-238
View File
@@ -578,138 +578,6 @@ async fn mid_turn_compact_success_broadcasts_start_and_done() {
assert_eq!(new_id_in_event, Some(worker.segment_id()));
}
/// Regression: `Worker::compact()` must reset the in-memory
/// `extract_pointer` so extract keeps firing on the new compacted
/// session.
///
/// Without the reset, the pointer's `processed_through_history_len`
/// holds the old (typically large) item count, while the new compacted
/// session starts with a much shorter history (`[summary, ...]`).
/// `cumulative_input_tokens_since` would then filter every new
/// usage record out (their `history_len` is below the stale pointer)
/// and extract would never re-fire for the rest of the process.
const EXTRACT_PLUS_COMPACT_MANIFEST: &str = r#"
[worker]
name = "test-worker"
pwd = "./"
[model]
scheme = "anthropic"
model_id = "test-model"
[engine]
max_tokens = 100
[memory]
workspace_id = "test-workspace"
settings_revision = 1
language = "English"
extract_threshold = 1
[compaction]
compact_threshold = 1
compact_retained_tokens = 0
[[scope.allow]]
target = "./"
permission = "write"
"#;
fn finish_memory_extraction_tool_use_events(call_id: &str) -> Vec<LlmEvent> {
let input = serde_json::json!({
"staged_count": 0,
"no_candidates_reason": "test run has no durable candidates"
})
.to_string();
vec![
LlmEvent::tool_use_start(0, call_id, "FinishMemoryExtraction"),
LlmEvent::tool_input_delta(0, input),
LlmEvent::tool_use_stop(0),
LlmEvent::Status(StatusEvent {
status: ResponseStatus::Completed,
}),
]
}
#[tokio::test]
async fn compact_resets_extract_pointer_so_extract_can_fire_again() {
// Mock LLM responses, in call order:
// [0] first run with usage(1000) so extract threshold (=1) fires.
// [1] extract worker invokes FinishMemoryExtraction with empty output.
// [2] extract worker closes after the tool result.
// [3] compact worker invokes write_summary.
// [4] compact worker closes after the tool result.
let client = MockClient::new(vec![
text_events_with_usage("hi", 1000),
finish_memory_extraction_tool_use_events("ec1"),
single_text_events("done"),
write_summary_tool_use_events("sc1", "summary"),
single_text_events("done"),
]);
let mut worker = make_worker_with_manifest(EXTRACT_PLUS_COMPACT_MANIFEST, client).await;
worker.run_text("first").await.unwrap();
// extract fires; pointer becomes Some.
worker.try_post_run_extract().await.unwrap();
assert!(
worker.extract_pointer().is_some(),
"extract_pointer should be Some after a successful extract"
);
// Compact runs. Without the fix the in-memory pointer would still
// reference the old Segment's history_len.
worker.try_pre_run_compact().await;
assert!(
worker.extract_pointer().is_none(),
"extract_pointer must be reset to None after compact (matches cold-restore on the new Segment)"
);
}
/// `extract_threshold = 0` is treated as "disabled" — without this, a
/// raw `>=` comparison against `tokens_since` would fire extract on
/// every post-run regardless of activity. Mirrors the consolidation
/// zero-threshold convention so users have a single way to opt out
/// without removing the `[memory]` section.
const EXTRACT_THRESHOLD_ZERO_MANIFEST: &str = r#"
[worker]
name = "test-worker"
pwd = "./"
[model]
scheme = "anthropic"
model_id = "test-model"
[engine]
max_tokens = 100
[memory]
extract_threshold = 0
[[scope.allow]]
target = "./"
permission = "write"
"#;
#[tokio::test]
async fn extract_threshold_zero_is_disabled() {
// Mock provides exactly one response — the first run. If extract
// were treated as "fire on any change" because of `tokens_since >= 0`,
// it would call into the extract worker and exhaust the mock.
let client = MockClient::new(vec![text_events_with_usage("hi", 1000)]);
let mut worker = make_worker_with_manifest(EXTRACT_THRESHOLD_ZERO_MANIFEST, client).await;
worker.run_text("first").await.unwrap();
worker
.try_post_run_extract()
.await
.expect("extract_threshold=0 must skip silently, not fail");
assert!(
worker.extract_pointer().is_none(),
"no extract should have run — pointer must remain None"
);
}
#[tokio::test]
async fn pre_run_compact_failure_broadcasts_start_and_failed() {
// Only the first run has a response. Compaction will run the
@@ -746,112 +614,6 @@ async fn pre_run_compact_failure_broadcasts_start_and_failed() {
);
}
// ---------------------------------------------------------------------------
// Detached post-run memory jobs (`spawn_post_run_memory_jobs` /
// `wait_for_memory_jobs`). Covers the detach round-trip and the structural
// invariant that the cloned memory-task Worker shares `SegmentState` with the
// source Worker, so that `save_extension` from the background extract does not
// leave the next turn's `save_user_input` looking at a stale session pointer.
const EXTRACT_NO_COMPACT_MANIFEST: &str = r#"
[worker]
name = "test-worker"
pwd = "./"
[model]
scheme = "anthropic"
model_id = "test-model"
[engine]
max_tokens = 100
[memory]
workspace_id = "test-workspace"
settings_revision = 1
language = "English"
extract_threshold = 1
[[scope.allow]]
target = "./"
permission = "write"
"#;
#[tokio::test]
async fn extract_large_unprocessed_range_does_not_abort_on_input_occupancy() {
let client = MockClient::new(vec![
text_events_with_usage("recorded", 1000),
finish_memory_extraction_tool_use_events("ec-large"),
single_text_events("done"),
]);
let mut worker = make_worker_with_manifest(EXTRACT_NO_COMPACT_MANIFEST, client).await;
let large_request = format!("remember this large slice: {}", "x ".repeat(200_000));
worker.run_text(&large_request).await.unwrap();
worker.try_post_run_extract().await.expect(
"large unprocessed extract ranges must reach the extract worker, not abort locally",
);
assert!(
worker.extract_pointer().is_some(),
"successful extract should advance the pointer even when the input range is large"
);
}
#[tokio::test]
async fn spawn_and_wait_drives_extract_to_completion() {
let client = MockClient::new(vec![
text_events_with_usage("hi", 1000),
finish_memory_extraction_tool_use_events("ec1"),
single_text_events("done"),
]);
let mut worker = make_worker_with_manifest(EXTRACT_NO_COMPACT_MANIFEST, client).await;
worker.run_text("first").await.unwrap();
assert!(
worker.extract_pointer().is_none(),
"extract has not run yet — pointer must be None"
);
worker.spawn_post_run_memory_jobs();
worker.wait_for_memory_jobs().await;
assert!(
worker.extract_pointer().is_some(),
"spawn + wait must complete extract; pointer should be set"
);
}
#[tokio::test]
async fn detached_extract_does_not_fork_session_log() {
// Source worker and the cloned memory-task worker share `SegmentState` via
// `Arc<_>`. The detached extract advances the entry tally through
// `save_extension`; the next `run` must see that same tally so
// `ensure_head_or_fork` does not spawn a new session.
let client = MockClient::new(vec![
text_events_with_usage("hi", 1000),
finish_memory_extraction_tool_use_events("ec1"),
single_text_events("done"),
text_events_with_usage("ok", 1000),
]);
let mut worker = make_worker_with_manifest(EXTRACT_NO_COMPACT_MANIFEST, client).await;
worker.run_text("first").await.unwrap();
let session_before = worker.segment_id();
worker.spawn_post_run_memory_jobs();
worker.wait_for_memory_jobs().await;
worker.run_text("second").await.unwrap();
let session_after = worker.segment_id();
assert_eq!(
session_before, session_after,
"detached extract's save_extension and the next turn's save_user_input \
must share the entry tally through SegmentState a fork here means the \
clone carried its own counter"
);
}
#[tokio::test]
async fn controller_compact_method_emits_start_and_done() {
let client = MockClient::new(vec![
+4
View File
@@ -38,6 +38,10 @@ required-features = ["typescript"]
name = "generate_memory_api_types"
required-features = ["typescript"]
[[example]]
name = "generate_skill_api_types"
required-features = ["typescript"]
[[example]]
name = "generate_auth_api_types"
required-features = ["typescript"]
@@ -0,0 +1,3 @@
fn main() {
print!("{}", workspace_api::skill_api_typescript());
}
+602
View File
@@ -1934,6 +1934,408 @@ pub struct RepositoryAccessProjection {
pub bindings: Vec<RepositorySshAccessBinding>,
}
pub const SKILL_CATALOG_AUTHORITY: &str = "workspace-config-skills-v1";
pub const SKILL_API_MAX_SAFE_INTEGER: u64 = 9_007_199_254_740_991;
pub const SKILL_API_MAX_CATALOG_ENTRIES: usize = 500;
pub const SKILL_API_MAX_OVERRIDES: usize = 64;
pub const SKILL_API_MAX_DIAGNOSTICS: usize = 100;
pub const SKILL_API_MAX_RESOURCES: usize = 500;
pub const SKILL_API_MAX_ALLOWED_TOOLS: usize = 100;
pub const SKILL_API_MAX_NAME_BYTES: usize = 128;
pub const SKILL_API_MAX_LABEL_BYTES: usize = 4_096;
pub const SKILL_API_MAX_BODY_BYTES: usize = 1_048_576;
pub const SKILL_API_MAX_PATH_BYTES: usize = 1_024;
pub const SKILL_API_MAX_DIGEST_BYTES: usize = 128;
pub const SKILL_API_MAX_RESPONSE_BYTES: usize = 2_097_152;
#[derive(Debug, Clone, Copy, Serialize, Deserialize, PartialEq, Eq)]
#[cfg_attr(feature = "typescript", derive(ts_rs::TS))]
#[cfg_attr(feature = "typescript", ts(rename_all = "snake_case"))]
#[serde(rename_all = "snake_case")]
pub enum SkillDiagnosticSeverity {
Error,
Warning,
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
#[cfg_attr(feature = "typescript", derive(ts_rs::TS))]
#[serde(deny_unknown_fields)]
pub struct SkillDiagnostic {
pub severity: SkillDiagnosticSeverity,
pub code: String,
pub message: String,
#[serde(default, skip_serializing_if = "Option::is_none")]
#[cfg_attr(feature = "typescript", ts(optional))]
pub source: Option<String>,
}
impl SkillDiagnostic {
pub fn error(
code: impl Into<String>,
message: impl Into<String>,
source: Option<String>,
) -> Self {
Self {
severity: SkillDiagnosticSeverity::Error,
code: code.into(),
message: message.into(),
source,
}
}
pub fn warning(
code: impl Into<String>,
message: impl Into<String>,
source: Option<String>,
) -> Self {
Self {
severity: SkillDiagnosticSeverity::Warning,
code: code.into(),
message: message.into(),
source,
}
}
}
#[derive(Debug, Clone, Copy, Serialize, Deserialize, PartialEq, Eq)]
#[cfg_attr(feature = "typescript", derive(ts_rs::TS))]
#[cfg_attr(feature = "typescript", ts(rename_all = "snake_case"))]
#[serde(rename_all = "snake_case")]
pub enum SkillSourceKind {
Builtin,
Workspace,
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
#[cfg_attr(feature = "typescript", derive(ts_rs::TS))]
#[serde(deny_unknown_fields)]
pub struct SkillProvenance {
pub kind: SkillSourceKind,
pub id: String,
#[serde(default, skip_serializing_if = "Option::is_none")]
#[cfg_attr(feature = "typescript", ts(optional))]
pub virtual_path: Option<String>,
#[serde(default, skip_serializing_if = "Option::is_none")]
#[cfg_attr(feature = "typescript", ts(optional, type = "number"))]
pub revision: Option<u64>,
#[serde(default, skip_serializing_if = "Option::is_none")]
#[cfg_attr(feature = "typescript", ts(optional))]
pub source_digest: Option<String>,
#[serde(default, skip_serializing_if = "Option::is_none")]
#[cfg_attr(feature = "typescript", ts(optional))]
pub tree_digest: Option<String>,
}
#[derive(Debug, Clone, Copy, Serialize, Deserialize, PartialEq, Eq)]
#[cfg_attr(feature = "typescript", derive(ts_rs::TS))]
#[cfg_attr(feature = "typescript", ts(rename_all = "snake_case"))]
#[serde(rename_all = "snake_case")]
pub enum SkillActivationStatus {
Active,
Inactive,
}
#[derive(Debug, Clone, Copy, Serialize, Deserialize, PartialEq, Eq)]
#[cfg_attr(feature = "typescript", derive(ts_rs::TS))]
#[cfg_attr(feature = "typescript", ts(rename_all = "snake_case"))]
#[serde(rename_all = "snake_case")]
pub enum SkillProjectionStatus {
Valid,
Invalid,
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
#[cfg_attr(feature = "typescript", derive(ts_rs::TS))]
#[serde(deny_unknown_fields)]
pub struct SkillProjectionIdentity {
#[cfg_attr(feature = "typescript", ts(type = "number"))]
pub config_revision: u64,
pub tree_digest: String,
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
#[cfg_attr(feature = "typescript", derive(ts_rs::TS))]
#[serde(deny_unknown_fields)]
pub struct SkillResourceRef {
pub kind: String,
pub name: String,
pub supported: bool,
#[serde(default, skip_serializing_if = "Option::is_none")]
#[cfg_attr(feature = "typescript", ts(optional))]
pub diagnostic: Option<String>,
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
#[cfg_attr(feature = "typescript", derive(ts_rs::TS))]
#[serde(deny_unknown_fields)]
pub struct SkillCatalogEntry {
pub name: String,
pub description: String,
pub activation_status: SkillActivationStatus,
pub projection_status: SkillProjectionStatus,
pub provenance: SkillProvenance,
pub overrides: Vec<SkillProvenance>,
pub diagnostics: Vec<SkillDiagnostic>,
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
#[cfg_attr(feature = "typescript", derive(ts_rs::TS))]
#[serde(deny_unknown_fields)]
pub struct SkillCatalogResponse {
pub authority: String,
pub projection: SkillProjectionIdentity,
pub entries: Vec<SkillCatalogEntry>,
pub diagnostics: Vec<SkillDiagnostic>,
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
#[cfg_attr(feature = "typescript", derive(ts_rs::TS))]
#[serde(deny_unknown_fields)]
pub struct SkillDetailResponse {
pub authority: String,
pub projection: SkillProjectionIdentity,
pub name: String,
pub description: String,
pub provenance: SkillProvenance,
pub overrides: Vec<SkillProvenance>,
pub diagnostics: Vec<SkillDiagnostic>,
pub activation_status: SkillActivationStatus,
pub projection_status: SkillProjectionStatus,
pub body: String,
pub allowed_tools: Vec<String>,
pub allowed_tools_status: String,
pub resources: Vec<SkillResourceRef>,
}
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum SkillApiValidationError {
CollectionTooLarge,
StringTooLarge,
InvalidProjectionIdentity,
InvalidProvenance,
InvalidVirtualPath,
StaleProjection,
}
impl std::fmt::Display for SkillApiValidationError {
fn fmt(&self, formatter: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
let message = match self {
Self::CollectionTooLarge => "Skill API collection exceeds its limit",
Self::StringTooLarge => "Skill API string exceeds its limit",
Self::InvalidProjectionIdentity => "Skill API projection identity is invalid",
Self::InvalidProvenance => "Skill API provenance is invalid",
Self::InvalidVirtualPath => "Skill API virtual path is invalid",
Self::StaleProjection => "Workspace Skill projection is stale",
};
formatter.write_str(message)
}
}
impl std::error::Error for SkillApiValidationError {}
impl SkillProjectionIdentity {
fn validate(&self) -> Result<(), SkillApiValidationError> {
validate_safe_integer(self.config_revision)?;
validate_nonempty_string(&self.tree_digest, SKILL_API_MAX_DIGEST_BYTES)
.map_err(|_| SkillApiValidationError::InvalidProjectionIdentity)
}
}
impl SkillProvenance {
fn validate(
&self,
projection: &SkillProjectionIdentity,
) -> Result<(), SkillApiValidationError> {
validate_nonempty_string(&self.id, SKILL_API_MAX_LABEL_BYTES)?;
validate_optional_string(self.virtual_path.as_deref(), SKILL_API_MAX_PATH_BYTES)?;
validate_optional_string(self.source_digest.as_deref(), SKILL_API_MAX_DIGEST_BYTES)?;
validate_optional_string(self.tree_digest.as_deref(), SKILL_API_MAX_DIGEST_BYTES)?;
if let Some(revision) = self.revision {
validate_safe_integer(revision)?;
}
let expected_prefix = match self.kind {
SkillSourceKind::Builtin => "builtin:",
SkillSourceKind::Workspace => "workspace:",
};
if !self.id.starts_with(expected_prefix)
|| self
.virtual_path
.as_deref()
.is_none_or(|path| !is_virtual_path(path))
|| self.source_digest.is_none()
{
return Err(SkillApiValidationError::InvalidProvenance);
}
match self.kind {
SkillSourceKind::Builtin => {
if self.revision.is_some() || self.tree_digest.is_some() {
return Err(SkillApiValidationError::InvalidProvenance);
}
}
SkillSourceKind::Workspace => {
let Some(revision) = self.revision else {
return Err(SkillApiValidationError::InvalidProvenance);
};
let Some(tree_digest) = self.tree_digest.as_deref() else {
return Err(SkillApiValidationError::InvalidProvenance);
};
if revision != projection.config_revision || tree_digest != projection.tree_digest {
return Err(SkillApiValidationError::StaleProjection);
}
}
}
Ok(())
}
}
impl SkillCatalogEntry {
fn validate(
&self,
projection: &SkillProjectionIdentity,
) -> Result<(), SkillApiValidationError> {
validate_nonempty_string(&self.name, SKILL_API_MAX_NAME_BYTES)?;
validate_string(&self.description, SKILL_API_MAX_LABEL_BYTES)?;
validate_collection(&self.overrides, SKILL_API_MAX_OVERRIDES)?;
validate_diagnostics(&self.diagnostics)?;
self.provenance.validate(projection)?;
for provenance in &self.overrides {
provenance.validate(projection)?;
}
Ok(())
}
}
impl SkillCatalogResponse {
pub fn validate(&self) -> Result<(), SkillApiValidationError> {
validate_nonempty_string(&self.authority, SKILL_API_MAX_LABEL_BYTES)?;
if self.authority != SKILL_CATALOG_AUTHORITY {
return Err(SkillApiValidationError::InvalidProjectionIdentity);
}
self.projection.validate()?;
validate_collection(&self.entries, SKILL_API_MAX_CATALOG_ENTRIES)?;
validate_diagnostics(&self.diagnostics)?;
for entry in &self.entries {
entry.validate(&self.projection)?;
}
Ok(())
}
}
impl SkillDetailResponse {
pub fn validate(&self) -> Result<(), SkillApiValidationError> {
validate_nonempty_string(&self.authority, SKILL_API_MAX_LABEL_BYTES)?;
if self.authority != SKILL_CATALOG_AUTHORITY {
return Err(SkillApiValidationError::InvalidProjectionIdentity);
}
self.projection.validate()?;
validate_nonempty_string(&self.name, SKILL_API_MAX_NAME_BYTES)?;
validate_string(&self.description, SKILL_API_MAX_LABEL_BYTES)?;
validate_string(&self.body, SKILL_API_MAX_BODY_BYTES)?;
validate_strings(
&self.allowed_tools,
SKILL_API_MAX_ALLOWED_TOOLS,
SKILL_API_MAX_LABEL_BYTES,
)?;
validate_nonempty_string(&self.allowed_tools_status, SKILL_API_MAX_LABEL_BYTES)?;
validate_resources(&self.resources)?;
validate_collection(&self.overrides, SKILL_API_MAX_OVERRIDES)?;
validate_diagnostics(&self.diagnostics)?;
self.provenance.validate(&self.projection)?;
for provenance in &self.overrides {
provenance.validate(&self.projection)?;
}
Ok(())
}
}
fn validate_safe_integer(value: u64) -> Result<(), SkillApiValidationError> {
if value <= SKILL_API_MAX_SAFE_INTEGER {
Ok(())
} else {
Err(SkillApiValidationError::InvalidProjectionIdentity)
}
}
fn validate_collection<T>(values: &[T], limit: usize) -> Result<(), SkillApiValidationError> {
if values.len() <= limit {
Ok(())
} else {
Err(SkillApiValidationError::CollectionTooLarge)
}
}
fn validate_string(value: &str, limit: usize) -> Result<(), SkillApiValidationError> {
if value.len() <= limit {
Ok(())
} else {
Err(SkillApiValidationError::StringTooLarge)
}
}
fn validate_nonempty_string(value: &str, limit: usize) -> Result<(), SkillApiValidationError> {
validate_string(value, limit)?;
if value.is_empty() {
Err(SkillApiValidationError::StringTooLarge)
} else {
Ok(())
}
}
fn validate_optional_string(
value: Option<&str>,
limit: usize,
) -> Result<(), SkillApiValidationError> {
if let Some(value) = value {
validate_nonempty_string(value, limit)?;
}
Ok(())
}
fn validate_strings(
values: &[String],
collection_limit: usize,
string_limit: usize,
) -> Result<(), SkillApiValidationError> {
validate_collection(values, collection_limit)?;
for value in values {
validate_nonempty_string(value, string_limit)?;
}
Ok(())
}
fn validate_resources(resources: &[SkillResourceRef]) -> Result<(), SkillApiValidationError> {
validate_collection(resources, SKILL_API_MAX_RESOURCES)?;
for resource in resources {
validate_nonempty_string(&resource.kind, SKILL_API_MAX_LABEL_BYTES)?;
validate_nonempty_string(&resource.name, SKILL_API_MAX_PATH_BYTES)?;
if !is_virtual_path(&resource.name) {
return Err(SkillApiValidationError::InvalidVirtualPath);
}
validate_optional_string(resource.diagnostic.as_deref(), SKILL_API_MAX_LABEL_BYTES)?;
}
Ok(())
}
fn is_virtual_path(value: &str) -> bool {
!value.starts_with('/')
&& !value.contains('\\')
&& value
.split('/')
.all(|component| !component.is_empty() && component != "." && component != "..")
}
fn validate_diagnostics(diagnostics: &[SkillDiagnostic]) -> Result<(), SkillApiValidationError> {
validate_collection(diagnostics, SKILL_API_MAX_DIAGNOSTICS)?;
for diagnostic in diagnostics {
validate_nonempty_string(&diagnostic.code, SKILL_API_MAX_LABEL_BYTES)?;
validate_nonempty_string(&diagnostic.message, SKILL_API_MAX_LABEL_BYTES)?;
validate_optional_string(diagnostic.source.as_deref(), SKILL_API_MAX_PATH_BYTES)?;
}
Ok(())
}
#[cfg(feature = "typescript")]
pub fn catalog_typescript() -> String {
use ts_rs::TS;
@@ -2005,6 +2407,37 @@ pub fn repository_access_api_typescript() -> String {
)
}
#[cfg(feature = "typescript")]
pub fn skill_api_typescript() -> String {
use ts_rs::TS;
let config = ts_rs::Config::default();
let declarations = [
SkillDiagnosticSeverity::decl(&config),
SkillDiagnostic::decl(&config),
SkillSourceKind::decl(&config),
SkillProvenance::decl(&config),
SkillActivationStatus::decl(&config),
SkillProjectionStatus::decl(&config),
SkillProjectionIdentity::decl(&config),
SkillResourceRef::decl(&config),
SkillCatalogEntry::decl(&config),
SkillCatalogResponse::decl(&config),
SkillDetailResponse::decl(&config),
];
let limits = format!(
"export const SKILL_API_AUTHORITY = \"{SKILL_CATALOG_AUTHORITY}\" as const;\n\nexport const SKILL_API_LIMITS = {{\n maxSafeInteger: {SKILL_API_MAX_SAFE_INTEGER},\n maxCatalogEntries: {SKILL_API_MAX_CATALOG_ENTRIES},\n maxOverrides: {SKILL_API_MAX_OVERRIDES},\n maxDiagnostics: {SKILL_API_MAX_DIAGNOSTICS},\n maxResources: {SKILL_API_MAX_RESOURCES},\n maxAllowedTools: {SKILL_API_MAX_ALLOWED_TOOLS},\n maxNameBytes: {SKILL_API_MAX_NAME_BYTES},\n maxLabelBytes: {SKILL_API_MAX_LABEL_BYTES},\n maxBodyBytes: {SKILL_API_MAX_BODY_BYTES},\n maxPathBytes: {SKILL_API_MAX_PATH_BYTES},\n maxDigestBytes: {SKILL_API_MAX_DIGEST_BYTES},\n maxResponseBytes: {SKILL_API_MAX_RESPONSE_BYTES},\n}} as const;"
);
format!(
"// Generated from workspace-api. Do not edit by hand.\n// Regenerate: cargo run -q -p workspace-api --features typescript --example generate_skill_api_types > web/workspace/src/lib/generated/skill-api.ts\n\n{limits}\n\n{}\n",
declarations
.into_iter()
.map(|declaration| format!("export {declaration}"))
.collect::<Vec<_>>()
.join("\n\n")
)
}
#[cfg(feature = "typescript")]
pub fn auth_api_typescript() -> String {
use ts_rs::TS;
@@ -2172,6 +2605,36 @@ mod memory_typescript_tests {
}
}
#[cfg(all(test, feature = "typescript"))]
mod skill_typescript_tests {
#[test]
fn generated_skill_api_contract_is_current() {
let expected = super::skill_api_typescript();
let path = std::path::Path::new(env!("CARGO_MANIFEST_DIR"))
.join("../../web/workspace/src/lib/generated/skill-api.ts");
let actual = std::fs::read_to_string(&path)
.unwrap_or_else(|error| panic!("failed to read {}: {error}", path.display()));
assert_eq!(
normalize(&actual),
normalize(&expected),
"regenerate Skill API TypeScript types with `cargo run -q -p workspace-api --features typescript --example generate_skill_api_types > web/workspace/src/lib/generated/skill-api.ts` and format the generated file",
);
}
fn normalize(value: &str) -> String {
value
.chars()
.filter_map(|character| match character {
character if character.is_whitespace() => None,
',' => Some(';'),
character => Some(character),
})
.collect::<String>()
.replace("=|", "=")
.replace(";}", "}")
}
}
#[cfg(all(test, feature = "typescript"))]
mod workdir_typescript_tests {
#[test]
@@ -2205,6 +2668,145 @@ mod workdir_typescript_tests {
mod tests {
use super::*;
fn skill_projection() -> SkillProjectionIdentity {
SkillProjectionIdentity {
config_revision: 42,
tree_digest: "tree-digest".to_string(),
}
}
fn builtin_skill_provenance() -> SkillProvenance {
SkillProvenance {
kind: SkillSourceKind::Builtin,
id: "builtin:errors".to_string(),
virtual_path: Some("skills/errors/SKILL.md".to_string()),
revision: None,
source_digest: Some("builtin-source-digest".to_string()),
tree_digest: None,
}
}
fn workspace_skill_provenance() -> SkillProvenance {
SkillProvenance {
kind: SkillSourceKind::Workspace,
id: "workspace:skills/release/SKILL.md".to_string(),
virtual_path: Some("skills/release/SKILL.md".to_string()),
revision: Some(42),
source_digest: Some("workspace-source-digest".to_string()),
tree_digest: Some("tree-digest".to_string()),
}
}
#[test]
fn skill_catalog_round_trips_builtin_workspace_and_invalid_projection_entries() {
let response = SkillCatalogResponse {
authority: "workspace-config-skills-v1".to_string(),
projection: skill_projection(),
entries: vec![
SkillCatalogEntry {
name: "errors".to_string(),
description: "Builtin guidance".to_string(),
activation_status: SkillActivationStatus::Active,
projection_status: SkillProjectionStatus::Valid,
provenance: builtin_skill_provenance(),
overrides: vec![],
diagnostics: vec![],
},
SkillCatalogEntry {
name: "release".to_string(),
description: "Workspace guidance".to_string(),
activation_status: SkillActivationStatus::Inactive,
projection_status: SkillProjectionStatus::Invalid,
provenance: workspace_skill_provenance(),
overrides: vec![builtin_skill_provenance()],
diagnostics: vec![SkillDiagnostic {
severity: SkillDiagnosticSeverity::Error,
code: "invalid_projection".to_string(),
message: "invalid projected Skill".to_string(),
source: Some("skills/release/SKILL.md".to_string()),
}],
},
],
diagnostics: vec![],
};
response.validate().expect("fixture should be valid");
let json = serde_json::to_string(&response).expect("serialize Skill catalog");
let decoded: SkillCatalogResponse =
serde_json::from_str(&json).expect("deserialize Skill catalog");
assert_eq!(decoded, response);
assert!(!json.contains("\"revision\":null"));
assert!(!json.contains("\"tree_digest\":null"));
}
#[test]
fn skill_detail_round_trips_shared_response() {
let response = SkillDetailResponse {
authority: "workspace-config-skills-v1".to_string(),
projection: skill_projection(),
name: "release".to_string(),
description: "Workspace guidance".to_string(),
body: "# Release\n".to_string(),
allowed_tools: vec!["Bash".to_string()],
allowed_tools_status: "experimental_hint_only".to_string(),
resources: vec![],
activation_status: SkillActivationStatus::Active,
projection_status: SkillProjectionStatus::Valid,
provenance: workspace_skill_provenance(),
overrides: vec![],
diagnostics: vec![],
};
response.validate().expect("fixture should be valid");
let decoded: SkillDetailResponse = serde_json::from_value(
serde_json::to_value(&response).expect("serialize Skill detail"),
)
.expect("deserialize Skill detail");
assert_eq!(decoded, response);
}
#[test]
fn skill_projection_validation_detects_stale_workspace_revision() {
let mut provenance = workspace_skill_provenance();
provenance.revision = Some(41);
let response = SkillCatalogResponse {
authority: "workspace-config-skills-v1".to_string(),
projection: skill_projection(),
entries: vec![SkillCatalogEntry {
name: "release".to_string(),
description: String::new(),
activation_status: SkillActivationStatus::Active,
projection_status: SkillProjectionStatus::Valid,
provenance,
overrides: vec![],
diagnostics: vec![],
}],
diagnostics: vec![],
};
assert_eq!(
response.validate(),
Err(SkillApiValidationError::StaleProjection)
);
}
#[test]
fn skill_dto_rejects_unknown_fields_and_unknown_provenance_kind() {
let unknown_field = serde_json::json!({
"authority": "workspace-config-skills-v1",
"projection": {"config_revision": 42, "tree_digest": "tree-digest"},
"entries": [],
"diagnostics": [],
"body": "must not be accepted"
});
assert!(serde_json::from_value::<SkillCatalogResponse>(unknown_field).is_err());
let mut provenance =
serde_json::to_value(workspace_skill_provenance()).expect("serialize provenance");
provenance["kind"] = serde_json::Value::String("newer_source_kind".to_string());
assert!(serde_json::from_value::<SkillProvenance>(provenance).is_err());
}
#[test]
fn memory_evidence_origins_round_trip_as_typed_provenance() {
let kinds = [
+16 -17
View File
@@ -7993,17 +7993,15 @@ fn start_memory_staging_consolidation(
total_bytes,
});
}
let reached_files = operation
.threshold_files
.is_some_and(|threshold| candidate_count >= threshold);
let reached_bytes = operation
.threshold_bytes
.is_some_and(|threshold| total_bytes >= threshold);
const CONSOLIDATION_THRESHOLD_FILES: usize = 5;
const CONSOLIDATION_THRESHOLD_BYTES: u64 = 50_000;
let reached_files = candidate_count >= CONSOLIDATION_THRESHOLD_FILES;
let reached_bytes = total_bytes >= CONSOLIDATION_THRESHOLD_BYTES;
if !operation.force && !reached_files && !reached_bytes {
return Ok(MemoryConsolidationOutput {
status: "skipped_below_threshold".to_string(),
summary: format!(
"Memory staging backlog has {candidate_count} candidate(s), {total_bytes} byte(s), below configured threshold."
"Memory staging backlog has {candidate_count} candidate(s), {total_bytes} byte(s), below Backend policy threshold."
),
candidate_count,
total_bytes,
@@ -19921,11 +19919,7 @@ mod tests {
let output = match start_memory_staging_consolidation(
api,
MemoryConsolidateStagingOperation {
force: true,
threshold_files: None,
threshold_bytes: None,
},
MemoryConsolidateStagingOperation { force: true },
) {
Ok(output) => output,
Err(_) => panic!("unexpected ApiError from memory consolidation trigger"),
@@ -19956,6 +19950,15 @@ mod tests {
)
.unwrap();
let below_threshold = start_memory_staging_consolidation(
api.clone(),
MemoryConsolidateStagingOperation { force: false },
)
.unwrap();
assert_eq!(below_threshold.status, "skipped_below_threshold");
assert_eq!(below_threshold.candidate_count, 1);
assert!(below_threshold.summary.contains("Backend policy threshold"));
let resolved_config_bundle = None;
let existing = api
.runtime
@@ -20010,11 +20013,7 @@ mod tests {
let second = match start_memory_staging_consolidation(
api.clone(),
MemoryConsolidateStagingOperation {
force: true,
threshold_files: None,
threshold_bytes: None,
},
MemoryConsolidateStagingOperation { force: true },
) {
Ok(output) => output,
Err(_) => panic!("unexpected ApiError from second memory consolidation trigger"),
+59 -8
View File
@@ -4,9 +4,11 @@ use config_source::{
ConfigSchemaContribution, MarkdownDocumentProjection, VirtualPath, project_markdown_document,
};
use serde::Deserialize;
use worker::skill::{
SkillActivationResponse, SkillCatalogEntry, SkillCatalogResponse, SkillDetailResponse,
SkillDiagnostic, SkillDiagnosticSeverity, SkillProvenance, SkillResourceRef, SkillSourceKind,
use worker::skill::SkillActivationResponse;
use workspace_api::{
SKILL_CATALOG_AUTHORITY, SkillActivationStatus, SkillCatalogEntry, SkillCatalogResponse,
SkillDetailResponse, SkillDiagnostic, SkillDiagnosticSeverity, SkillProjectionIdentity,
SkillProjectionStatus, SkillProvenance, SkillResourceRef, SkillSourceKind,
};
use crate::config_source::{
@@ -19,7 +21,6 @@ const BUILTIN_SKILL_VIRTUAL_PATH: &str = "builtin/skills/agent-skills/SKILL.md";
const SKILL_SCHEMA_PROVIDER_ID: &str = "builtin:skills";
const SKILL_SCHEMA_NAMESPACE: &str = "skills";
const SKILL_SCHEMA_VERSION: &str = "1";
const SKILL_CATALOG_AUTHORITY: &str = "workspace-config-skills-v1";
/// Skill documents are values imported from `SKILL.md`. Known Agent Skills
/// frontmatter is typed while extension keys remain concrete values.
@@ -101,32 +102,54 @@ pub fn catalog(state: &WorkspaceConfigState) -> Result<SkillCatalogResponse, Ski
.into_values()
.map(|skill| skill.catalog_entry())
.collect();
Ok(SkillCatalogResponse {
let response = SkillCatalogResponse {
authority: SKILL_CATALOG_AUTHORITY.to_string(),
projection: projection_identity(state),
entries,
diagnostics: Vec::new(),
})
};
response
.validate()
.map_err(|error| SkillError::InvalidProjection(error.to_string()))?;
Ok(response)
}
pub fn lint(state: &WorkspaceConfigState) -> Result<SkillCatalogResponse, SkillError> {
catalog(state)
}
fn projection_identity(state: &WorkspaceConfigState) -> SkillProjectionIdentity {
SkillProjectionIdentity {
config_revision: state.snapshot.revision,
tree_digest: state.snapshot.digest.clone(),
}
}
pub fn detail(state: &WorkspaceConfigState, name: &str) -> Result<SkillDetailResponse, SkillError> {
let skill = merged_skills(state)?
.remove(name)
.ok_or_else(|| SkillError::NotFound(name.to_string()))?;
Ok(SkillDetailResponse {
let activation_status = skill.activation_status();
let projection_status = skill.projection_status();
let response = SkillDetailResponse {
authority: SKILL_CATALOG_AUTHORITY.to_string(),
projection: projection_identity(state),
name: skill.name,
description: skill.description,
provenance: skill.provenance,
overrides: skill.overrides,
diagnostics: skill.diagnostics,
activation_status,
projection_status,
body: skill.body,
allowed_tools: skill.allowed_tools,
allowed_tools_status: "experimental_hint_only".to_string(),
resources: skill.resources,
})
};
response
.validate()
.map_err(|error| SkillError::InvalidProjection(error.to_string()))?;
Ok(response)
}
pub fn activation(
@@ -415,10 +438,28 @@ impl ParsedSkill {
.any(|diagnostic| diagnostic.severity == SkillDiagnosticSeverity::Error)
}
fn activation_status(&self) -> SkillActivationStatus {
if self.has_errors() {
SkillActivationStatus::Inactive
} else {
SkillActivationStatus::Active
}
}
fn projection_status(&self) -> SkillProjectionStatus {
if self.has_errors() {
SkillProjectionStatus::Invalid
} else {
SkillProjectionStatus::Valid
}
}
fn catalog_entry(&self) -> SkillCatalogEntry {
SkillCatalogEntry {
name: self.name.clone(),
description: self.description.clone(),
activation_status: self.activation_status(),
projection_status: self.projection_status(),
provenance: self.provenance.clone(),
overrides: self.overrides.clone(),
diagnostics: self.diagnostics.clone(),
@@ -513,12 +554,20 @@ mod tests {
.unwrap();
assert_eq!(item.provenance.kind, SkillSourceKind::Workspace);
assert_eq!(item.provenance.revision, Some(9));
assert_eq!(catalog.projection.config_revision, 9);
assert_eq!(catalog.projection.tree_digest, state.snapshot.digest);
assert_eq!(item.activation_status, SkillActivationStatus::Active);
assert_eq!(item.projection_status, SkillProjectionStatus::Valid);
assert!(
item.diagnostics
.iter()
.all(|diagnostic| diagnostic.severity != SkillDiagnosticSeverity::Error)
);
let detail = detail(&state, "debug-rust").unwrap();
assert_eq!(detail.authority, SKILL_CATALOG_AUTHORITY);
assert_eq!(detail.projection.config_revision, 9);
assert_eq!(detail.activation_status, SkillActivationStatus::Active);
assert_eq!(detail.projection_status, SkillProjectionStatus::Valid);
assert_eq!(detail.body, "# Debug Rust\n");
assert_eq!(detail.allowed_tools, vec!["Read", "Grep"]);
assert_eq!(
@@ -579,6 +628,8 @@ mod tests {
.into_iter()
.find(|item| item.name == "debug-rust")
.unwrap();
assert_eq!(item.activation_status, SkillActivationStatus::Inactive);
assert_eq!(item.projection_status, SkillProjectionStatus::Invalid);
assert!(
item.diagnostics
.iter()
+27 -41
View File
@@ -222,55 +222,41 @@ permission = "write"
# # ref = "anthropic/claude-haiku-4-5"
# ===== [memory] =============================================================
# Memory subsystem の opt-in。
# - セクションが *ある* … memory tools (MemoryRead/Write/Edit) を登録、
# `<workspace>/memory/` と `<workspace>/`
# の通常 write を Worker 自体に対して deny する
# - セクションが *無い* … 何も起きない (legacy 動作)。
# `[memory]` だけ書いて中身を省略するのも有効 (全フィールド既定値で有効化)
# [memory]
# ===== [feature.memory] ======================================================
# Memory は `feature.memory` だけを入口にする。resolved Worker Manifest では
# Profile由来の設定を `profile` に、Backend由来のWorkspace設定snapshotを
# `workspace_settings` に分離して保存する。`workspace_settings` はBackendだけが
# bindする信頼済み入力で、Profile・Browser・model入力から指定できない
# `profile.enabled = false` の場合、Memory tools、Feature prompt contributionによる
# resident injection、extract、consolidation requestをすべて無効にし、snapshotも保持しない
#
# # 任意。デフォルト: Worker の pwd (構築時)。
# # 必ず絶対パス (相対なら manifest base 起点で resolve)。
# workspace_root = "/abs/path/to/workspace"
# [feature.memory.profile]
# enabled = true
# staging_tools = false
#
# # 任意。デフォルト: tool 側既定 = 20。
# # MemoryQuery / MemoryQuery が 1 回に返す最大件数。
# query_result_limit = 20
# [feature.memory.profile.resident]
# inject_summary = true
#
# # 任意。デフォルト: tool 側既定 = 3。
# # 各マッチ前後に表示するコンテキスト行数。`query` 省略時は無視。
# query_excerpt_lines = 3
# [feature.memory.profile.extraction]
# enabled = true
# threshold = 30000
# worker_max_turns = 8
#
# # 任意。デフォルト: メインモデルを `clone_boxed()` で複製
# # extract ワーカーのモデル ([model] と同じ形式)。
# # Haiku / 4o-mini / Flash クラスの軽量 reasoning モデル推奨。
# # [memory.extract_model]
# # 任意。省略時はmain modelをcloneする
# # [feature.memory.profile.extraction.model]
# # ref = "anthropic/claude-haiku-4-5"
#
# # 任意。デフォルト: なし (extract 自動発火を完全停止)。
# # 前回 extract pointer 以降の累積入力 token がこの値を超えると extract 起動。
# # ※ memory tools と resident injection は extract_threshold が None でも動く。
# extract_threshold = 30000
# [feature.memory.profile.consolidation]
# request_enabled = true
#
# # 任意。デフォルト: 8 (`defaults::MEMORY_EXTRACT_WORKER_MAX_TURNS`)
# # extract worker 自身の tool loop 上限。Rust config で None の場合のみ無制限。
# extract_worker_max_turns = 8
# # Backendがresolved Manifestへbindする。手書き/Profile入力では指定しない
# # [feature.memory.workspace_settings]
# # workspace_id = "workspace-id"
# # settings_revision = 1
# # language = "日本語"
#
# # 任意。デフォルト: メインモデルを `clone_boxed()` で複製。
# # consolidation ワーカーのモデル。reasoning クラス推奨
# # [memory.consolidation_model]
# # ref = "anthropic/claude-sonnet-4-6"
#
# # 任意。デフォルト: なし。
# # `_staging/` のエントリ数がこの値以上で consolidation 発火 (files / bytes は OR)。
# consolidation_threshold_files = 50
#
# # 任意。デフォルト: なし。
# # `_staging/` の総バイト数がこの値以上で consolidation 発火 (files / bytes は OR)。
# # files / bytes の両方が None だと consolidation 完全無効。
# consolidation_threshold_bytes = 1048576
# Query結果/抜粋の上限とconsolidation eligibility/thresholdはoperation/Backend
# policyが所有し、通常Worker Manifestには含めない。legacy `[memory]` は拒否する
# ===== [skills] =============================================================
+9 -7
View File
@@ -23,7 +23,15 @@ compaction = {
feature = {
task = { enabled = true; };
memory = { enabled = true; };
memory = {
enabled = true;
resident = { inject_summary = true; };
extraction = {
enabled = true;
threshold = 50000;
};
consolidation = { request_enabled = true; };
};
web = { enabled = true; };
image = { enabled = true; };
sub_worker = { enabled = false; };
@@ -40,12 +48,6 @@ feature = {
};
};
memory = {
extract_threshold = 50000;
consolidation_threshold_files = 5;
consolidation_threshold_bytes = 50000;
};
web = {
enabled = true;
search = {
+1 -1
View File
@@ -6,7 +6,7 @@ import "./base.dcdl" // {
feature = {
task = { enabled = true; };
memory = { enabled = false; staging = false; };
memory = { enabled = false; staging_tools = false; };
web = { enabled = true; };
image = { enabled = true; };
sub_worker = { enabled = true; };
+1 -1
View File
@@ -5,7 +5,7 @@ import "./base.dcdl" // {
feature = {
task = { enabled = false; };
memory = { enabled = true; staging = true; };
memory = { enabled = true; staging_tools = true; };
web = { enabled = false; };
sub_worker = { enabled = false; };
worker = { enabled = false; };
+1 -1
View File
@@ -6,7 +6,7 @@
"dev": "deno run -A npm:vite@7.2.7 dev",
"dev:backend": "cd ../.. && cargo run -p yoi-workspace-server --bin yoi-server -- serve --listen 127.0.0.1:8787",
"check": "deno run -A npm:@sveltejs/kit@2.49.4 sync && deno run -A npm:svelte-check@4.3.4 --tsconfig ./tsconfig.json",
"test": "deno test --allow-read=src,test,tests --allow-env=LOG,VSCODE_TEXTMATE_DEBUG,NODE_ENV tests/workspace-model.test.ts tests/workspace-catalog.test.ts tests/profile-api.test.ts src/lib/workspace/auth/model.test.ts tests/auth-api.test.ts src/lib/workspace/api/http.test.ts src/lib/workspace/api/workers.test.ts src/lib/workspace/header/breadcrumb-model.test.ts src/lib/workspace/console/chat-submit.test.ts test/composer-history.test.ts tests/composer-paste.test.ts src/lib/workspace/console/composer-command.test.ts src/lib/workspace/console/composer-draft.test.ts src/lib/workspace/console/composer-completion.test.ts src/lib/workspace/console/markdown.test.ts test/console/ansi.test.ts src/lib/workspace/console/model.test.ts src/lib/workspace/companion/api.test.ts tests/workdir-api.test.ts src/lib/workspace/console/tasks.test.ts test/ticket-detail-route-reuse.test.ts test/repositories/ui.test.ts src/lib/workspace/console/worker-console.ui.test.ts src/lib/workspace/settings/model.test.ts src/lib/workspace/sidebar/override-stack.test.ts src/lib/workspace/sidebar/workers.test.ts src/lib/workspace/sidebar/workspace-switcher.test.ts src/lib/workspace/sidebar/worker-subscription.test.ts src/lib/workspace/sidebar/worker-launch.test.ts test/sidebar/worker-actions.test.ts src/lib/workspace/tickets/merge-request-resources.test.ts src/lib/workspace/tickets/ticket-panel.test.ts test/merge-request-status.test.ts test/config-source/decodal-grammar.test.ts test/config-source/editor-state.test.ts test/config-source/fixed-schema-wrapper.test.ts test/config-source/toolchain.test.ts test/config-source/wasm-parity.test.ts test/repository-access/api.test.ts test/repository-access/loader.test.ts test/repository-access/ui.test.ts",
"test": "deno test --allow-read=src,test,tests --allow-env=LOG,VSCODE_TEXTMATE_DEBUG,NODE_ENV tests/workspace-model.test.ts tests/workspace-catalog.test.ts tests/profile-api.test.ts tests/skill-api.test.ts src/lib/workspace/auth/model.test.ts tests/auth-api.test.ts src/lib/workspace/api/http.test.ts src/lib/workspace/api/workers.test.ts src/lib/workspace/header/breadcrumb-model.test.ts src/lib/workspace/console/chat-submit.test.ts test/composer-history.test.ts tests/composer-paste.test.ts src/lib/workspace/console/composer-command.test.ts src/lib/workspace/console/composer-draft.test.ts src/lib/workspace/console/composer-completion.test.ts src/lib/workspace/console/markdown.test.ts test/console/ansi.test.ts src/lib/workspace/console/model.test.ts src/lib/workspace/companion/api.test.ts tests/workdir-api.test.ts src/lib/workspace/console/tasks.test.ts test/ticket-detail-route-reuse.test.ts test/repositories/ui.test.ts src/lib/workspace/console/worker-console.ui.test.ts src/lib/workspace/settings/model.test.ts src/lib/workspace/sidebar/override-stack.test.ts src/lib/workspace/sidebar/workers.test.ts src/lib/workspace/sidebar/workspace-switcher.test.ts src/lib/workspace/sidebar/worker-subscription.test.ts src/lib/workspace/sidebar/worker-launch.test.ts test/sidebar/worker-actions.test.ts src/lib/workspace/tickets/merge-request-resources.test.ts src/lib/workspace/tickets/ticket-panel.test.ts test/merge-request-status.test.ts test/config-source/decodal-grammar.test.ts test/config-source/editor-state.test.ts test/config-source/fixed-schema-wrapper.test.ts test/config-source/toolchain.test.ts test/config-source/wasm-parity.test.ts test/repository-access/api.test.ts test/repository-access/loader.test.ts test/repository-access/ui.test.ts",
"build": "deno run -A npm:vite@7.2.7 build",
"preview": "deno run -A npm:vite@7.2.7 preview"
},
@@ -0,0 +1,88 @@
// Generated from workspace-api. Do not edit by hand.
// Regenerate: cargo run -q -p workspace-api --features typescript --example generate_skill_api_types > web/workspace/src/lib/generated/skill-api.ts
export const SKILL_API_AUTHORITY = "workspace-config-skills-v1" as const;
export const SKILL_API_LIMITS = {
maxSafeInteger: 9007199254740991,
maxCatalogEntries: 500,
maxOverrides: 64,
maxDiagnostics: 100,
maxResources: 500,
maxAllowedTools: 100,
maxNameBytes: 128,
maxLabelBytes: 4096,
maxBodyBytes: 1048576,
maxPathBytes: 1024,
maxDigestBytes: 128,
maxResponseBytes: 2097152,
} as const;
export type SkillDiagnosticSeverity = "error" | "warning";
export type SkillDiagnostic = {
severity: SkillDiagnosticSeverity;
code: string;
message: string;
source?: string;
};
export type SkillSourceKind = "builtin" | "workspace";
export type SkillProvenance = {
kind: SkillSourceKind;
id: string;
virtual_path?: string;
revision?: number;
source_digest?: string;
tree_digest?: string;
};
export type SkillActivationStatus = "active" | "inactive";
export type SkillProjectionStatus = "valid" | "invalid";
export type SkillProjectionIdentity = {
config_revision: number;
tree_digest: string;
};
export type SkillResourceRef = {
kind: string;
name: string;
supported: boolean;
diagnostic?: string;
};
export type SkillCatalogEntry = {
name: string;
description: string;
activation_status: SkillActivationStatus;
projection_status: SkillProjectionStatus;
provenance: SkillProvenance;
overrides: Array<SkillProvenance>;
diagnostics: Array<SkillDiagnostic>;
};
export type SkillCatalogResponse = {
authority: string;
projection: SkillProjectionIdentity;
entries: Array<SkillCatalogEntry>;
diagnostics: Array<SkillDiagnostic>;
};
export type SkillDetailResponse = {
authority: string;
projection: SkillProjectionIdentity;
name: string;
description: string;
provenance: SkillProvenance;
overrides: Array<SkillProvenance>;
diagnostics: Array<SkillDiagnostic>;
activation_status: SkillActivationStatus;
projection_status: SkillProjectionStatus;
body: string;
allowed_tools: Array<string>;
allowed_tools_status: string;
resources: Array<SkillResourceRef>;
};
@@ -1,11 +1,13 @@
import {
loadWorkspaceSkillCatalog,
loadWorkspaceSkillDetail,
workspaceApiPath,
workspaceRoute,
workspaceSkillActivationPath,
workspaceSkillCatalogPath,
workspaceSkillDetailPath,
} from "./http.ts";
import { SKILL_API_LIMITS } from "$lib/generated/skill-api.ts";
declare const Deno: {
test(name: string, fn: () => Promise<void> | void): void;
@@ -89,11 +91,24 @@ Deno.test("loadWorkspaceSkillCatalog fetches lightweight catalog", async () => {
return Promise.resolve(
new Response(
JSON.stringify({
authority: "workspace-backend-skills-v0",
authority: "workspace-config-skills-v1",
projection: {
config_revision: 7,
tree_digest: "tree-digest",
},
entries: [{
name: "triage-errors",
description: "Use when triaging errors.",
provenance: { kind: "workspace", id: "workspace:triage-errors" },
activation_status: "active",
projection_status: "valid",
provenance: {
kind: "workspace",
id: "workspace:triage-errors",
virtual_path: "skills/triage-errors/SKILL.md",
revision: 7,
source_digest: "source-digest",
tree_digest: "tree-digest",
},
overrides: [],
diagnostics: [],
}],
@@ -110,3 +125,44 @@ Deno.test("loadWorkspaceSkillCatalog fetches lightweight catalog", async () => {
assertEquals(result.data?.entries[0].name, "triage-errors");
assertEquals(JSON.stringify(result.data).includes("SKILL.md body"), false);
});
Deno.test("Skill loaders redact and bound non-success response diagnostics", async () => {
const secret = "SENSITIVE-SKILL-BODY-CONTENT".repeat(300);
const result = await loadWorkspaceSkillCatalog(
(() =>
Promise.resolve(new Response(secret, { status: 500 }))) as typeof fetch,
"ws-1",
);
assertEquals(result.data, null);
assertEquals(result.error, "Skill API request failed with HTTP 500");
assert(
!result.error?.includes(secret.slice(0, 64)),
"Skill API diagnostic must not expose response body content",
);
assert(
(result.error?.length ?? 0) <= 256,
"Skill API diagnostic must remain bounded",
);
});
Deno.test("Skill loaders stop reading success responses above the wire byte limit", async () => {
const oversized = `{"body":"${
"x".repeat(SKILL_API_LIMITS.maxResponseBytes + 1)
}"}`;
const result = await loadWorkspaceSkillDetail(
(() =>
Promise.resolve(
new Response(oversized, { status: 200 }),
)) as typeof fetch,
"ws-1",
"release",
);
assertEquals(result.data, null);
assertEquals(result.error, "Skill API response exceeds its byte limit");
assert(
!result.error?.includes(oversized.slice(0, 64)),
"Skill API diagnostic must not expose oversized response content",
);
});
+89 -46
View File
@@ -1,58 +1,32 @@
import { SKILL_API_LIMITS } from "$lib/generated/skill-api.ts";
import type {
SkillCatalogResponse,
SkillDetailResponse,
} from "$lib/generated/skill-api.ts";
import {
parseSkillCatalogResponse,
parseSkillDetailResponse,
SkillApiContractError,
} from "$lib/workspace/skills/api.ts";
export type ApiResult<T> = {
data: T | null;
error: string | null;
};
export type SkillDiagnosticSeverity = "error" | "warning";
export type { SkillCatalogResponse, SkillDetailResponse };
export type SkillDiagnostic = {
severity: SkillDiagnosticSeverity;
code: string;
message: string;
source?: string;
type JsonLoadPolicy = {
diagnosticLabel: string;
maxResponseBytes: number;
};
export type SkillProvenance = {
kind: "builtin" | "workspace";
id: string;
virtual_path?: string;
revision?: number;
source_digest?: string;
tree_digest?: string;
const SKILL_API_LOAD_POLICY: JsonLoadPolicy = {
diagnosticLabel: "Skill API",
maxResponseBytes: SKILL_API_LIMITS.maxResponseBytes,
};
export type SkillCatalogEntry = {
name: string;
description: string;
provenance: SkillProvenance;
overrides: SkillProvenance[];
diagnostics: SkillDiagnostic[];
};
export type SkillCatalogResponse = {
authority: string;
entries: SkillCatalogEntry[];
diagnostics: SkillDiagnostic[];
};
export type SkillResourceRef = {
kind: string;
name: string;
supported: boolean;
diagnostic?: string;
};
export type SkillDetailResponse = {
name: string;
description: string;
provenance: SkillProvenance;
overrides: SkillProvenance[];
diagnostics: SkillDiagnostic[];
body: string;
allowed_tools: string[];
allowed_tools_status: string;
resources: SkillResourceRef[];
};
class ResponseByteLimitError extends Error {}
function normalizePath(path: string): string {
if (!path || path === "/") return "";
@@ -95,6 +69,9 @@ export async function loadWorkspaceSkillCatalog(
return loadJson<SkillCatalogResponse>(
fetchFn,
workspaceSkillCatalogPath(workspaceId),
undefined,
parseSkillCatalogResponse,
SKILL_API_LOAD_POLICY,
);
}
@@ -106,6 +83,9 @@ export async function loadWorkspaceSkillDetail(
return loadJson<SkillDetailResponse>(
fetchFn,
workspaceSkillDetailPath(workspaceId, name),
undefined,
parseSkillDetailResponse,
SKILL_API_LOAD_POLICY,
);
}
@@ -114,19 +94,38 @@ export async function loadJson<T>(
path: string,
init?: RequestInit,
parse: (value: unknown) => T = (value) => value as T,
policy?: JsonLoadPolicy,
): Promise<ApiResult<T>> {
try {
const response = await fetchFn(path, init);
if (!response.ok) {
if (policy) {
await response.body?.cancel();
return {
data: null,
error:
`${policy.diagnosticLabel} request failed with HTTP ${response.status}`,
};
}
const text = await response.text();
return {
data: null,
error: text || `${path} request failed (${response.status})`,
};
}
const payload: unknown = await response.json();
const payload: unknown = policy
? await readBoundedJson(response, policy.maxResponseBytes)
: await response.json();
return { data: parse(payload), error: null };
} catch (error) {
if (policy) {
const diagnostic = error instanceof SkillApiContractError
? error.message
: error instanceof ResponseByteLimitError
? `${policy.diagnosticLabel} response exceeds its byte limit`
: `${policy.diagnosticLabel} response is invalid`;
return { data: null, error: diagnostic.slice(0, 256) };
}
return {
data: null,
error: error instanceof Error ? error.message : `${path} request failed`,
@@ -134,6 +133,50 @@ export async function loadJson<T>(
}
}
async function readBoundedJson(
response: Response,
maxBytes: number,
): Promise<unknown> {
const contentLength = response.headers.get("content-length");
if (contentLength !== null) {
const parsedLength = Number(contentLength);
if (Number.isFinite(parsedLength) && parsedLength > maxBytes) {
await response.body?.cancel();
throw new ResponseByteLimitError();
}
}
const reader = response.body?.getReader();
if (!reader) {
throw new Error("response body is unavailable");
}
const chunks: Uint8Array[] = [];
let totalBytes = 0;
try {
while (true) {
const { done, value } = await reader.read();
if (done) break;
totalBytes += value.byteLength;
if (totalBytes > maxBytes) {
await reader.cancel();
throw new ResponseByteLimitError();
}
chunks.push(value);
}
} finally {
reader.releaseLock();
}
const bytes = new Uint8Array(totalBytes);
let offset = 0;
for (const chunk of chunks) {
bytes.set(chunk, offset);
offset += chunk.byteLength;
}
const text = new TextDecoder("utf-8", { fatal: true }).decode(bytes);
return JSON.parse(text) as unknown;
}
async function requireJson<T>(response: Response, path: string): Promise<T> {
if (!response.ok) {
const text = await response.text();
@@ -0,0 +1,455 @@
import {
SKILL_API_AUTHORITY,
SKILL_API_LIMITS,
type SkillActivationStatus,
type SkillCatalogEntry,
type SkillCatalogResponse,
type SkillDetailResponse,
type SkillDiagnostic,
type SkillDiagnosticSeverity,
type SkillProjectionIdentity,
type SkillProjectionStatus,
type SkillProvenance,
type SkillResourceRef,
type SkillSourceKind,
} from "$lib/generated/skill-api.ts";
export class SkillApiContractError extends Error {
constructor(message: string) {
super(message);
this.name = "SkillApiContractError";
}
}
export function parseSkillCatalogResponse(
value: unknown,
): SkillCatalogResponse {
const record = strictObject(value, [
"authority",
"projection",
"entries",
"diagnostics",
], "Skill catalog response");
const authority = boundedString(
record.authority,
"Skill catalog authority",
SKILL_API_LIMITS.maxLabelBytes,
false,
);
if (authority !== SKILL_API_AUTHORITY) {
throw contractError("unsupported Skill catalog authority");
}
const projection = parseProjection(record.projection);
return {
authority,
projection,
entries: boundedArray(
record.entries,
"Skill catalog entries",
SKILL_API_LIMITS.maxCatalogEntries,
).map((entry) => parseCatalogEntry(entry, projection)),
diagnostics: parseDiagnostics(record.diagnostics),
};
}
export function parseSkillDetailResponse(value: unknown): SkillDetailResponse {
const record = strictObject(value, [
"authority",
"projection",
"name",
"description",
"provenance",
"overrides",
"diagnostics",
"activation_status",
"projection_status",
"body",
"allowed_tools",
"allowed_tools_status",
"resources",
], "Skill detail response");
const authority = boundedString(
record.authority,
"Skill detail authority",
SKILL_API_LIMITS.maxLabelBytes,
false,
);
if (authority !== SKILL_API_AUTHORITY) {
throw contractError("unsupported Skill detail authority");
}
const projection = parseProjection(record.projection);
return {
authority,
projection,
name: boundedString(
record.name,
"Skill name",
SKILL_API_LIMITS.maxNameBytes,
false,
),
description: boundedString(
record.description,
"Skill description",
SKILL_API_LIMITS.maxLabelBytes,
true,
),
provenance: parseProvenance(record.provenance, projection),
overrides: parseProvenances(record.overrides, projection),
diagnostics: parseDiagnostics(record.diagnostics),
activation_status: activationStatus(record.activation_status),
projection_status: projectionStatus(record.projection_status),
body: boundedString(
record.body,
"Skill body",
SKILL_API_LIMITS.maxBodyBytes,
true,
),
allowed_tools: boundedArray(
record.allowed_tools,
"Skill allowed tools",
SKILL_API_LIMITS.maxAllowedTools,
).map((tool) =>
boundedString(
tool,
"Skill allowed tool",
SKILL_API_LIMITS.maxLabelBytes,
false,
)
),
allowed_tools_status: boundedString(
record.allowed_tools_status,
"Skill allowed-tools status",
SKILL_API_LIMITS.maxLabelBytes,
false,
),
resources: boundedArray(
record.resources,
"Skill resources",
SKILL_API_LIMITS.maxResources,
).map(parseResource),
};
}
function parseCatalogEntry(
value: unknown,
projection: SkillProjectionIdentity,
): SkillCatalogEntry {
const record = strictObject(value, [
"name",
"description",
"activation_status",
"projection_status",
"provenance",
"overrides",
"diagnostics",
], "Skill catalog entry");
return {
name: boundedString(
record.name,
"Skill name",
SKILL_API_LIMITS.maxNameBytes,
false,
),
description: boundedString(
record.description,
"Skill description",
SKILL_API_LIMITS.maxLabelBytes,
true,
),
activation_status: activationStatus(record.activation_status),
projection_status: projectionStatus(record.projection_status),
provenance: parseProvenance(record.provenance, projection),
overrides: parseProvenances(record.overrides, projection),
diagnostics: parseDiagnostics(record.diagnostics),
};
}
function parseProjection(value: unknown): SkillProjectionIdentity {
const record = strictObject(
value,
["config_revision", "tree_digest"],
"Skill projection identity",
);
return {
config_revision: safeInteger(
record.config_revision,
"Skill config revision",
),
tree_digest: boundedString(
record.tree_digest,
"Skill tree digest",
SKILL_API_LIMITS.maxDigestBytes,
false,
),
};
}
function parseProvenances(
value: unknown,
projection: SkillProjectionIdentity,
): SkillProvenance[] {
return boundedArray(
value,
"Skill overrides",
SKILL_API_LIMITS.maxOverrides,
).map((provenance) => parseProvenance(provenance, projection));
}
function parseProvenance(
value: unknown,
projection: SkillProjectionIdentity,
): SkillProvenance {
const record = strictObject(
value,
[
"kind",
"id",
"virtual_path",
"revision",
"source_digest",
"tree_digest",
],
"Skill provenance",
[
"virtual_path",
"revision",
"source_digest",
"tree_digest",
],
);
const kind = sourceKind(record.kind);
const id = boundedString(
record.id,
"Skill provenance id",
SKILL_API_LIMITS.maxLabelBytes,
false,
);
const virtualPath = optionalBoundedString(
record.virtual_path,
"Skill virtual path",
SKILL_API_LIMITS.maxPathBytes,
);
const sourceDigest = optionalBoundedString(
record.source_digest,
"Skill source digest",
SKILL_API_LIMITS.maxDigestBytes,
);
const treeDigest = optionalBoundedString(
record.tree_digest,
"Skill provenance tree digest",
SKILL_API_LIMITS.maxDigestBytes,
);
const revision = record.revision === undefined
? undefined
: safeInteger(record.revision, "Skill provenance revision");
if (
!id.startsWith(`${kind}:`) || virtualPath === undefined ||
sourceDigest === undefined || !isVirtualPath(virtualPath)
) {
throw contractError("invalid Skill provenance");
}
if (kind === "builtin") {
if (revision !== undefined || treeDigest !== undefined) {
throw contractError("invalid built-in Skill provenance");
}
} else {
if (revision === undefined || treeDigest === undefined) {
throw contractError("incomplete Workspace Skill provenance");
}
if (
revision !== projection.config_revision ||
treeDigest !== projection.tree_digest
) {
throw contractError("stale Workspace Skill projection");
}
}
return {
kind,
id,
virtual_path: virtualPath,
revision,
source_digest: sourceDigest,
tree_digest: treeDigest,
};
}
function parseDiagnostics(value: unknown): SkillDiagnostic[] {
return boundedArray(
value,
"Skill diagnostics",
SKILL_API_LIMITS.maxDiagnostics,
).map((diagnostic) => {
const record = strictObject(
diagnostic,
[
"severity",
"code",
"message",
"source",
],
"Skill diagnostic",
["source"],
);
return {
severity: diagnosticSeverity(record.severity),
code: boundedString(
record.code,
"Skill diagnostic code",
SKILL_API_LIMITS.maxLabelBytes,
false,
),
message: boundedString(
record.message,
"Skill diagnostic message",
SKILL_API_LIMITS.maxLabelBytes,
false,
),
source: optionalBoundedString(
record.source,
"Skill diagnostic source",
SKILL_API_LIMITS.maxPathBytes,
),
};
});
}
function parseResource(value: unknown): SkillResourceRef {
const record = strictObject(
value,
[
"kind",
"name",
"supported",
"diagnostic",
],
"Skill resource",
["diagnostic"],
);
if (typeof record.supported !== "boolean") {
throw contractError("Skill resource supported must be a boolean");
}
const name = boundedString(
record.name,
"Skill resource name",
SKILL_API_LIMITS.maxPathBytes,
false,
);
if (!isVirtualPath(name)) {
throw contractError("invalid Skill resource virtual path");
}
return {
kind: boundedString(
record.kind,
"Skill resource kind",
SKILL_API_LIMITS.maxLabelBytes,
false,
),
name,
supported: record.supported,
diagnostic: optionalBoundedString(
record.diagnostic,
"Skill resource diagnostic",
SKILL_API_LIMITS.maxLabelBytes,
),
};
}
function sourceKind(value: unknown): SkillSourceKind {
if (value === "builtin" || value === "workspace") return value;
throw contractError("unsupported Skill provenance kind");
}
function diagnosticSeverity(value: unknown): SkillDiagnosticSeverity {
if (value === "error" || value === "warning") return value;
throw contractError("unsupported Skill diagnostic severity");
}
function activationStatus(value: unknown): SkillActivationStatus {
if (value === "active" || value === "inactive") return value;
throw contractError("unsupported Skill activation status");
}
function projectionStatus(value: unknown): SkillProjectionStatus {
if (value === "valid" || value === "invalid") return value;
throw contractError("unsupported Skill projection status");
}
function safeInteger(value: unknown, label: string): number {
if (
typeof value !== "number" || !Number.isSafeInteger(value) || value < 0 ||
value > SKILL_API_LIMITS.maxSafeInteger
) {
throw contractError(`${label} must be a non-negative safe integer`);
}
return value;
}
function boundedArray(
value: unknown,
label: string,
limit: number,
): unknown[] {
if (!Array.isArray(value) || value.length > limit) {
throw contractError(`${label} must be a bounded array`);
}
return value;
}
function optionalBoundedString(
value: unknown,
label: string,
limit: number,
): string | undefined {
return value === undefined
? undefined
: boundedString(value, label, limit, false);
}
function boundedString(
value: unknown,
label: string,
limit: number,
allowEmpty: boolean,
): string {
if (
typeof value !== "string" || (!allowEmpty && value.length === 0) ||
new TextEncoder().encode(value).length > limit
) {
throw contractError(`${label} must be a bounded string`);
}
return value;
}
function isVirtualPath(value: string): boolean {
return !value.startsWith("/") && !value.includes("\\") &&
value.split("/").every((part) =>
part !== "" && part !== "." && part !== ".."
);
}
function strictObject(
value: unknown,
allowedKeys: readonly string[],
label: string,
optionalKeys: readonly string[] = [],
): Record<string, unknown> {
if (typeof value !== "object" || value === null || Array.isArray(value)) {
throw contractError(`${label} must be an object`);
}
const record = value as Record<string, unknown>;
const allowed = new Set(allowedKeys);
if (Object.keys(record).some((key) => !allowed.has(key))) {
throw contractError(`${label} contains unknown fields`);
}
const optional = new Set(optionalKeys);
if (allowedKeys.some((key) => !optional.has(key) && !(key in record))) {
throw contractError(`${label} is missing required fields`);
}
return record;
}
function contractError(message: string): SkillApiContractError {
return new SkillApiContractError(message.slice(0, 256));
}
+221
View File
@@ -0,0 +1,221 @@
import {
parseSkillCatalogResponse,
parseSkillDetailResponse,
SkillApiContractError,
} from "../src/lib/workspace/skills/api.ts";
import { SKILL_API_LIMITS } from "../src/lib/generated/skill-api.ts";
declare const Deno: {
test(name: string, fn: () => Promise<void> | void): void;
};
function assert(condition: unknown, message: string): asserts condition {
if (!condition) throw new Error(message);
}
function assertEquals<T>(actual: T, expected: T): void {
if (JSON.stringify(actual) !== JSON.stringify(expected)) {
throw new Error(
`expected ${JSON.stringify(expected)}, got ${JSON.stringify(actual)}`,
);
}
}
function assertContractError(value: () => unknown, expected: string): void {
try {
value();
} catch (error) {
assert(
error instanceof SkillApiContractError,
"expected SkillApiContractError",
);
assert(
error.message.includes(expected),
`expected bounded diagnostic containing ${expected}, got ${error.message}`,
);
assert(error.message.length <= 256, "diagnostic must remain bounded");
return;
}
throw new Error("expected parser to reject malformed Skill response");
}
function builtinProvenance() {
return {
kind: "builtin",
id: "builtin:errors",
virtual_path: "skills/errors/SKILL.md",
source_digest: "builtin-source-digest",
};
}
function workspaceProvenance() {
return {
kind: "workspace",
id: "workspace:release",
virtual_path: "skills/release/SKILL.md",
revision: 42,
source_digest: "workspace-source-digest",
tree_digest: "tree-digest",
};
}
function catalogFixture(): Record<string, unknown> {
return {
authority: "workspace-config-skills-v1",
projection: { config_revision: 42, tree_digest: "tree-digest" },
entries: [{
name: "errors",
description: "Builtin guidance",
activation_status: "active",
projection_status: "valid",
provenance: builtinProvenance(),
overrides: [],
diagnostics: [],
}, {
name: "release",
description: "Workspace guidance",
activation_status: "inactive",
projection_status: "invalid",
provenance: workspaceProvenance(),
overrides: [builtinProvenance()],
diagnostics: [{
severity: "error",
code: "invalid_projection",
message: "invalid projected Skill",
source: "workspace:release",
}],
}],
diagnostics: [],
};
}
function detailFixture(): Record<string, unknown> {
return {
authority: "workspace-config-skills-v1",
projection: { config_revision: 42, tree_digest: "tree-digest" },
name: "release",
description: "Workspace guidance",
provenance: workspaceProvenance(),
overrides: [],
diagnostics: [],
activation_status: "active",
projection_status: "valid",
body: "# Release\n",
allowed_tools: ["Bash"],
allowed_tools_status: "experimental_hint_only",
resources: [{
kind: "reference",
name: "skills/release/references/checklist.md",
supported: true,
}],
};
}
Deno.test("Skill catalog parser accepts generated builtin, Workspace, and invalid projection shapes", () => {
const parsed = parseSkillCatalogResponse(catalogFixture());
assertEquals(parsed.entries.length, 2);
assertEquals(parsed.entries[0].provenance.kind, "builtin");
assertEquals(parsed.entries[1].activation_status, "inactive");
assertEquals(parsed.entries[1].projection_status, "invalid");
assertEquals(parsed.projection.config_revision, 42);
});
Deno.test("Skill detail parser preserves shared generated DTO fields", () => {
const parsed = parseSkillDetailResponse(detailFixture());
assertEquals(parsed.name, "release");
assertEquals(parsed.allowed_tools, ["Bash"]);
assertEquals(parsed.resources[0].supported, true);
});
Deno.test("Skill parser rejects stale Workspace projection revision and digest", () => {
const staleRevision = catalogFixture();
(staleRevision.projection as Record<string, unknown>).config_revision = 43;
assertContractError(
() => parseSkillCatalogResponse(staleRevision),
"stale Workspace Skill projection",
);
const staleDigest = catalogFixture();
(staleDigest.projection as Record<string, unknown>).tree_digest = "new-tree";
assertContractError(
() => parseSkillCatalogResponse(staleDigest),
"stale Workspace Skill projection",
);
});
Deno.test("Skill parser fails closed on unknown fields and newer enum values", () => {
const unknownField = catalogFixture();
unknownField.unexpected = true;
assertContractError(
() => parseSkillCatalogResponse(unknownField),
"unknown fields",
);
const newerProvenance = catalogFixture();
const entries = newerProvenance.entries as Record<string, unknown>[];
(entries[0].provenance as Record<string, unknown>).kind = "remote_catalog";
assertContractError(
() => parseSkillCatalogResponse(newerProvenance),
"unsupported Skill provenance kind",
);
const newerStatus = catalogFixture();
const newerEntries = newerStatus.entries as Record<string, unknown>[];
newerEntries[0].projection_status = "stale";
assertContractError(
() => parseSkillCatalogResponse(newerStatus),
"unsupported Skill projection status",
);
});
Deno.test("Skill parser rejects unsafe revisions and oversized collections or strings", () => {
const unsafeRevision = catalogFixture();
(unsafeRevision.projection as Record<string, unknown>).config_revision =
Number.MAX_SAFE_INTEGER + 1;
assertContractError(
() => parseSkillCatalogResponse(unsafeRevision),
"safe integer",
);
const oversizedCatalog = catalogFixture();
const firstEntry = (oversizedCatalog.entries as unknown[])[0];
oversizedCatalog.entries = Array.from(
{ length: SKILL_API_LIMITS.maxCatalogEntries + 1 },
() => firstEntry,
);
assertContractError(
() => parseSkillCatalogResponse(oversizedCatalog),
"bounded array",
);
const oversizedDetail = detailFixture();
oversizedDetail.body = "x".repeat(SKILL_API_LIMITS.maxBodyBytes + 1);
assertContractError(
() => parseSkillDetailResponse(oversizedDetail),
"bounded string",
);
});
Deno.test("Skill parser diagnostics never include rejected Skill body content", () => {
const secret = "SENSITIVE-SKILL-BODY-CONTENT";
const malformed = detailFixture();
malformed.body = secret;
malformed.provenance = {
...workspaceProvenance(),
kind: "newer_source_kind",
};
try {
parseSkillDetailResponse(malformed);
throw new Error("expected malformed provenance to fail");
} catch (error) {
assert(
error instanceof SkillApiContractError,
"expected SkillApiContractError",
);
assert(
!error.message.includes(secret),
"diagnostic leaked Skill body content",
);
assert(error.message.length <= 256, "diagnostic must remain bounded");
}
});