Files
yoi/crates/tools/tests/integration.rs

1318 lines
45 KiB
Rust

//! Cross-tool integration tests exercising `core_builtin_tools()` end-to-end.
//!
//! `ToolServerHandle::register_tool` / `flush_pending` are `pub(crate)` in
//! agen, so from here we exercise the factories directly — the same
//! code path that `flush_pending()` runs at production time.
use std::path::Path;
use std::sync::Arc;
use agen::tool::{
Tool, ToolDefinition, ToolError, ToolExecutionContext, ToolExecutionHandle,
ToolExecutionTerminal, ToolMeta,
};
use manifest::Scope;
use serde_json::json;
use tempfile::TempDir;
use tools::{
Tracker, core_builtin_tools, routed_builtin_tools, routed_view_image_tool, view_image_tool,
};
use workdir::{
LocalWorkdirSession, WorkdirAttachmentAlias, WorkdirPath, WorkdirSessionHandle,
WorkdirSessionRouter, WorkdirToolBroker, WorkdirToolScope, WorkdirToolScopePermission,
WorkdirToolScopeRule,
};
struct Registry {
entries: Vec<(ToolMeta, Arc<dyn Tool>)>,
}
impl Registry {
fn new(defs: Vec<ToolDefinition>) -> Self {
let entries = defs.into_iter().map(|f| f()).collect();
Self { entries }
}
fn get(&self, name: &str) -> Arc<dyn Tool> {
self.entries
.iter()
.find(|(m, _)| m.name == name)
.map(|(_, t)| Arc::clone(t))
.unwrap_or_else(|| panic!("tool not found: {name}"))
}
fn names(&self) -> Vec<&str> {
self.entries.iter().map(|(m, _)| m.name.as_str()).collect()
}
}
fn setup() -> (TempDir, Registry) {
let dir = TempDir::new().unwrap();
let scope = Scope::writable(dir.path()).unwrap();
let fs: WorkdirSessionHandle =
Arc::new(LocalWorkdirSession::new(scope, dir.path().to_path_buf()));
let tracker = Tracker::new();
let reg = Registry::new(core_builtin_tools(
fs,
tracker,
tools::bash_output::BashOutputView::new(),
));
(dir, reg)
}
fn setup_routed() -> (TempDir, TempDir, Arc<WorkdirSessionRouter>, Registry) {
let left = TempDir::new().unwrap();
let right = TempDir::new().unwrap();
let router = Arc::new(WorkdirSessionRouter::new());
for (alias, dir) in [("left", &left), ("right", &right)] {
let scope = Scope::writable(dir.path()).unwrap();
let session: WorkdirSessionHandle =
Arc::new(LocalWorkdirSession::new(scope, dir.path().to_path_buf()));
router
.attach(WorkdirAttachmentAlias::new(alias).unwrap(), session)
.unwrap();
}
let definitions = routed_builtin_tools(
router.clone(),
Tracker::new(),
tools::bash_output::BashOutputView::new(),
);
(left, right, router, Registry::new(definitions))
}
fn valid_gif_with_len(target_len: usize) -> Vec<u8> {
let fixture = include_bytes!("fixtures/view-image/valid.gif");
assert_eq!(
fixture.last(),
Some(&0x3b),
"GIF fixture must end in trailer"
);
let mut bytes = fixture[..fixture.len() - 1].to_vec();
bytes.extend_from_slice(&[0x21, 0xfe]);
let mut remaining = target_len
.checked_sub(bytes.len() + 2)
.expect("target must leave room for comment terminator and GIF trailer");
while remaining > 0 {
let mut block = remaining.min(256);
if remaining - block == 1 {
block -= 1;
}
assert!(block >= 2);
bytes.push((block - 1) as u8);
bytes.extend(std::iter::repeat_n(b'x', block - 1));
remaining -= block;
}
bytes.extend_from_slice(&[0, 0x3b]);
assert_eq!(bytes.len(), target_len);
bytes
}
async fn call(tool: &Arc<dyn Tool>, input: serde_json::Value) -> agen::tool::ToolOutput {
tool.execute(&input.to_string(), Default::default())
.await
.expect("tool execution failed")
}
async fn call_err(tool: &Arc<dyn Tool>, input: serde_json::Value) -> agen::tool::ToolError {
tool.execute(&input.to_string(), Default::default())
.await
.expect_err("expected error")
}
#[test]
fn core_builtin_tools_registers_full_set() {
let (_dir, reg) = setup();
let mut names = reg.names();
names.sort();
assert_eq!(names, vec!["Bash", "Edit", "Glob", "Grep", "Read", "Write"]);
}
#[test]
fn meta_has_description_and_schema() {
let (_dir, reg) = setup();
for (meta, _) in &reg.entries {
assert!(
!meta.description.is_empty(),
"{} missing description",
meta.name
);
// Input schema must be a JSON object
assert!(
meta.input_schema.is_object(),
"{} input_schema is not an object",
meta.name
);
}
}
#[test]
fn routed_tool_schemas_share_optional_target_workdir() {
let (_left, _right, router, mut reg) = setup_routed();
reg.entries.push(routed_view_image_tool(router)());
for (meta, _) in &reg.entries {
let target = &meta.input_schema["properties"]["target_workdir"];
assert!(target.is_object(), "{} lacks target_workdir", meta.name);
assert_eq!(target["type"], json!(["string", "null"]));
}
}
#[tokio::test]
async fn routed_tools_require_alias_only_when_attachment_set_is_ambiguous() {
let (_left, _right, router, reg) = setup_routed();
let error = call_err(&reg.get("Read"), json!({ "file_path": "same.txt" })).await;
let message = error.to_string();
assert!(message.contains("target_workdir_required"), "{message}");
assert!(message.contains("left"), "{message}");
assert!(message.contains("right"), "{message}");
let unknown = call_err(
&reg.get("Glob"),
json!({ "target_workdir": "missing", "pattern": "*" }),
)
.await
.to_string();
assert!(unknown.contains("unknown_target_workdir"), "{unknown}");
assert!(unknown.contains("available_aliases"), "{unknown}");
router.close_all().await.unwrap();
}
#[tokio::test]
async fn empty_routed_surface_returns_no_attachment_error() {
let router = Arc::new(WorkdirSessionRouter::new());
let reg = Registry::new(routed_builtin_tools(
router,
Tracker::new(),
tools::bash_output::BashOutputView::new(),
));
let message = call_err(&reg.get("Read"), json!({ "file_path": "file.txt" }))
.await
.to_string();
assert!(message.contains("no_workdir_attached"), "{message}");
}
#[tokio::test]
async fn target_alias_routes_each_tool_and_isolates_read_before_edit() {
let (left, right, router, mut reg) = setup_routed();
std::fs::write(left.path().join("same.txt"), "left value\n").unwrap();
std::fs::write(right.path().join("same.txt"), "right value\n").unwrap();
std::fs::write(left.path().join("only-left.log"), "LEFT-NEEDLE\n").unwrap();
std::fs::write(right.path().join("only-right.log"), "RIGHT-NEEDLE\n").unwrap();
let read = reg.get("Read");
let left_read = call(
&read,
json!({ "target_workdir": "left", "file_path": "same.txt" }),
)
.await;
assert!(left_read.content.unwrap().contains("left value"));
let edit = reg.get("Edit");
let isolated = call_err(
&edit,
json!({
"target_workdir": "right",
"file_path": "same.txt",
"old_string": "right",
"new_string": "edited"
}),
)
.await;
assert!(isolated.to_string().contains("has not been read"));
call(
&read,
json!({ "target_workdir": "right", "file_path": "same.txt" }),
)
.await;
call(
&edit,
json!({
"target_workdir": "right",
"file_path": "same.txt",
"old_string": "right",
"new_string": "edited"
}),
)
.await;
assert_eq!(
std::fs::read_to_string(right.path().join("same.txt")).unwrap(),
"edited value\n"
);
assert_eq!(
std::fs::read_to_string(left.path().join("same.txt")).unwrap(),
"left value\n"
);
let glob = call(
&reg.get("Glob"),
json!({ "target_workdir": "left", "pattern": "*.log" }),
)
.await;
assert!(glob.content.unwrap().contains("only-left.log"));
let grep = call(
&reg.get("Grep"),
json!({
"target_workdir": "right",
"pattern": "RIGHT-NEEDLE",
"output_mode": "content"
}),
)
.await;
assert!(grep.content.unwrap().contains("only-right.log"));
let bash = call(
&reg.get("Bash"),
json!({ "target_workdir": "right", "command": "pwd", "cwd": "." }),
)
.await;
assert_eq!(
std::fs::canonicalize(serde_json::from_str::<serde_json::Value>(&bash.content.unwrap()).unwrap()["stdout_preview"].as_str().unwrap().trim()).unwrap(),
std::fs::canonicalize(right.path()).unwrap()
);
let png = b"\x89PNG\r\n\x1a\nrouted";
std::fs::write(left.path().join("image.png"), png).unwrap();
reg.entries.push(routed_view_image_tool(router.clone())());
let image = call(
&reg.get("ViewImage"),
json!({ "target_workdir": "left", "path": "image.png" }),
)
.await;
let agen::tool::Attachment::Image(image) = &image.attachments[0];
assert_eq!(image.data(), png);
router.close_all().await.unwrap();
}
#[tokio::test]
async fn selected_attachment_capability_is_checked_without_fallback() {
let left = TempDir::new().unwrap();
let right = TempDir::new().unwrap();
let router = Arc::new(WorkdirSessionRouter::new());
router
.attach(
WorkdirAttachmentAlias::new("writable").unwrap(),
Arc::new(LocalWorkdirSession::new(
Scope::writable(left.path()).unwrap(),
left.path().to_path_buf(),
)),
)
.unwrap();
router
.attach(
WorkdirAttachmentAlias::new("readonly").unwrap(),
Arc::new(LocalWorkdirSession::materialized_bound(
workdir::Workdir::new("readonly"),
right.path().to_path_buf(),
right.path().to_path_buf(),
manifest::SharedScope::new(Scope::writable(right.path()).unwrap()),
workdir::WorkdirSessionCapabilities::READ_ONLY,
)),
)
.unwrap();
let reg = Registry::new(routed_builtin_tools(
router.clone(),
Tracker::new(),
tools::bash_output::BashOutputView::new(),
));
let message = call_err(
&reg.get("Write"),
json!({
"target_workdir": "readonly",
"file_path": "created.txt",
"content": "must not be routed elsewhere"
}),
)
.await
.to_string();
assert!(message.contains("does not support Write"), "{message}");
assert!(!left.path().join("created.txt").exists());
assert!(!right.path().join("created.txt").exists());
router.close_all().await.unwrap();
}
#[tokio::test]
async fn routed_tools_observe_detach_after_registration() {
let (_left, _right, router, reg) = setup_routed();
let alias = WorkdirAttachmentAlias::new("left").unwrap();
router.detach(&alias).await.unwrap();
let message = call_err(
&reg.get("Glob"),
json!({ "target_workdir": "left", "pattern": "*" }),
)
.await
.to_string();
assert!(message.contains("unknown_target_workdir"), "{message}");
assert!(message.contains("right"), "{message}");
router.close_all().await.unwrap();
}
#[tokio::test]
async fn detach_and_reattach_same_alias_requires_a_fresh_read() {
let first = TempDir::new().unwrap();
let second = TempDir::new().unwrap();
std::fs::write(first.path().join("same.txt"), "same bytes\n").unwrap();
std::fs::write(second.path().join("same.txt"), "same bytes\n").unwrap();
let router = Arc::new(WorkdirSessionRouter::new());
let alias = WorkdirAttachmentAlias::new("checkout").unwrap();
router
.attach(
alias.clone(),
Arc::new(LocalWorkdirSession::new(
Scope::writable(first.path()).unwrap(),
first.path().to_path_buf(),
)),
)
.unwrap();
let reg = Registry::new(routed_builtin_tools(
router.clone(),
Tracker::new(),
tools::bash_output::BashOutputView::new(),
));
call(
&reg.get("Read"),
json!({ "target_workdir": "checkout", "file_path": "same.txt" }),
)
.await;
router.detach(&alias).await.unwrap();
router
.attach(
alias.clone(),
Arc::new(LocalWorkdirSession::new(
Scope::writable(second.path()).unwrap(),
second.path().to_path_buf(),
)),
)
.unwrap();
let error = call_err(
&reg.get("Edit"),
json!({
"target_workdir": "checkout",
"file_path": "same.txt",
"old_string": "same",
"new_string": "changed"
}),
)
.await;
assert!(error.to_string().contains("has not been read"), "{error}");
assert_eq!(
std::fs::read_to_string(second.path().join("same.txt")).unwrap(),
"same bytes\n"
);
router.close_all().await.unwrap();
}
#[derive(Debug)]
struct ExternalValidatingSession {
inner: LocalWorkdirSession,
truncate_first_binary_response: bool,
command_response_losses: std::sync::atomic::AtomicUsize,
lose_terminal_only: bool,
mutate_after_first_binary_response: Option<std::path::PathBuf>,
}
#[async_trait::async_trait]
impl workdir::WorkdirSession for ExternalValidatingSession {
fn workdir(&self) -> &workdir::Workdir {
self.inner.workdir()
}
fn capabilities(&self) -> workdir::WorkdirSessionCapabilities {
self.inner.capabilities()
}
async fn stat(
&self,
request: workdir::StatRequest,
) -> Result<workdir::StatResult, workdir::WorkdirError> {
self.inner.stat(request).await
}
async fn read(
&self,
request: workdir::ReadRequest,
) -> Result<workdir::ReadResult, workdir::WorkdirError> {
let operation = workdir::http::WorkdirSessionOperation::Read(request);
let operation = workdir::external::ExternalWorkdirOperation::try_from(operation)
.map_err(workdir::WorkdirError::UnsupportedOperation)?;
let workdir::http::WorkdirSessionOperation::Read(request) = operation.into_inner() else {
unreachable!("validated Read changed operation kind")
};
self.inner.read(request).await
}
async fn read_bytes(
&self,
request: workdir::ReadBytesRequest,
) -> Result<workdir::ReadBytesResult, workdir::WorkdirError> {
let operation = workdir::http::WorkdirSessionOperation::ReadBytes(request);
let operation = workdir::external::ExternalWorkdirOperation::try_from(operation)
.map_err(workdir::WorkdirError::UnsupportedOperation)?;
let workdir::http::WorkdirSessionOperation::ReadBytes(request) = operation.into_inner()
else {
unreachable!("validated ReadBytes changed operation kind")
};
let result = self.inner.read_bytes(request).await?;
let result = workdir::external::ExternalWorkdirOperationResult::try_from(
workdir::http::WorkdirSessionOperationResult::ReadBytes(result),
)
.map_err(workdir::WorkdirError::UnsupportedOperation)?;
let workdir::http::WorkdirSessionOperationResult::ReadBytes(mut result) =
result.into_inner()
else {
unreachable!("validated ReadBytes result changed operation kind")
};
if result.offset == 0 {
if let Some(path) = &self.mutate_after_first_binary_response {
std::fs::write(path, vec![b'z'; result.total_bytes as usize]).unwrap();
}
if self.truncate_first_binary_response {
result.bytes.pop();
}
}
Ok(result)
}
async fn write(
&self,
request: workdir::WriteRequest,
) -> Result<workdir::WriteResult, workdir::WorkdirError> {
self.inner.write(request).await
}
async fn edit(
&self,
request: workdir::EditRequest,
) -> Result<workdir::EditResult, workdir::WorkdirError> {
self.inner.edit(request).await
}
async fn list(
&self,
request: workdir::ListRequest,
) -> Result<workdir::ListResult, workdir::WorkdirError> {
self.inner.list(request).await
}
async fn glob(
&self,
request: workdir::GlobRequest,
) -> Result<workdir::GlobResult, workdir::WorkdirError> {
self.inner.glob(request).await
}
async fn grep(
&self,
request: workdir::GrepRequest,
) -> Result<workdir::GrepResult, workdir::WorkdirError> {
self.inner.grep(request).await
}
async fn start_command(
&self,
request: workdir::CommandRequest,
) -> Result<workdir::CommandHandle, workdir::WorkdirError> {
self.inner.start_command(request).await
}
async fn command_status(
&self,
handle: workdir::CommandHandle,
) -> Result<workdir::CommandStatus, workdir::WorkdirError> {
self.inner.command_status(handle).await
}
async fn command_output(
&self,
request: workdir::CommandOutputRequest,
) -> Result<workdir::CommandOutput, workdir::WorkdirError> {
let output = self.inner.command_output(request).await?;
if (!self.lose_terminal_only || output.terminal.is_some())
&& self
.command_response_losses
.fetch_update(
std::sync::atomic::Ordering::AcqRel,
std::sync::atomic::Ordering::Acquire,
|remaining| remaining.checked_sub(1),
)
.is_ok()
{
// The production provider has committed a page, but its response is lost.
return Err(
workdir::WorkdirError::Transport("injected lost response".into())
.with_failure_reason(workdir::WorkdirFailureReason::HttpTimeout),
);
}
Ok(output)
}
async fn cancel_command(
&self,
handle: workdir::CommandHandle,
) -> Result<(), workdir::WorkdirError> {
self.inner.cancel_command(handle).await
}
async fn close(&self) -> Result<(), workdir::WorkdirError> {
self.inner.close().await
}
}
#[tokio::test]
async fn external_view_image_uses_a_protocol_bounded_request() {
let dir = TempDir::new().unwrap();
let scope = Scope::writable(dir.path()).unwrap();
let session: WorkdirSessionHandle = Arc::new(ExternalValidatingSession {
inner: LocalWorkdirSession::new(scope, dir.path().to_path_buf()),
truncate_first_binary_response: false,
command_response_losses: std::sync::atomic::AtomicUsize::new(0),
lose_terminal_only: false,
mutate_after_first_binary_response: None,
});
let png = include_bytes!("fixtures/view-image/valid.png");
assert!(png.contains(&0));
assert!(png.iter().any(|byte| *byte >= 0x80));
std::fs::write(dir.path().join("image.png"), png).unwrap();
let (_meta, tool) = view_image_tool(session)();
let output = call(&tool, json!({ "path": "image.png" })).await;
let agen::tool::Attachment::Image(image) = &output.attachments[0];
assert_eq!(image.mime_type(), "image/png");
assert_eq!(image.data(), png);
}
#[tokio::test]
async fn external_view_image_rejects_truncated_and_cross_version_chunks() {
let dir = TempDir::new().unwrap();
let image_path = dir.path().join("image.png");
let scope = Scope::writable(dir.path()).unwrap();
std::fs::write(&image_path, include_bytes!("fixtures/view-image/valid.png")).unwrap();
let truncated: WorkdirSessionHandle = Arc::new(ExternalValidatingSession {
inner: LocalWorkdirSession::new(scope.clone(), dir.path().to_path_buf()),
truncate_first_binary_response: true,
command_response_losses: std::sync::atomic::AtomicUsize::new(0),
lose_terminal_only: false,
mutate_after_first_binary_response: None,
});
let (_meta, tool) = view_image_tool(truncated)();
let error = call_err(&tool, json!({ "path": "image.png" }))
.await
.to_string();
assert!(
error.contains("truncated") || error.contains("partial"),
"{error}"
);
let large = valid_gif_with_len(1024 * 1024 + 32);
std::fs::write(&image_path, large).unwrap();
let changing: WorkdirSessionHandle = Arc::new(ExternalValidatingSession {
inner: LocalWorkdirSession::new(scope, dir.path().to_path_buf()),
truncate_first_binary_response: false,
command_response_losses: std::sync::atomic::AtomicUsize::new(0),
lose_terminal_only: false,
mutate_after_first_binary_response: Some(image_path),
});
let (_meta, tool) = view_image_tool(changing)();
let error = call_err(&tool, json!({ "path": "image.png" }))
.await
.to_string();
assert!(error.contains("changed"), "{error}");
}
#[tokio::test]
async fn view_image_enforces_the_ten_mib_boundary_without_partial_attachments() {
const LIMIT: usize = 10 * 1024 * 1024;
let dir = TempDir::new().unwrap();
let scope = Scope::writable(dir.path()).unwrap();
let session: WorkdirSessionHandle =
Arc::new(LocalWorkdirSession::new(scope, dir.path().to_path_buf()));
let (_meta, tool) = view_image_tool(session)();
let mut image = valid_gif_with_len(LIMIT);
std::fs::write(dir.path().join("boundary.gif"), &image).unwrap();
let output = call(&tool, json!({ "path": "boundary.gif" })).await;
let agen::tool::Attachment::Image(attachment) = &output.attachments[0];
assert_eq!(attachment.mime_type(), "image/gif");
assert_eq!(attachment.data(), image);
image.push(0);
std::fs::write(dir.path().join("boundary.gif"), image).unwrap();
let error = call_err(&tool, json!({ "path": "boundary.gif" }))
.await
.to_string();
assert!(error.contains("10485760-byte limit"), "{error}");
}
#[tokio::test]
async fn view_image_attaches_valid_supported_image_fixtures() {
let dir = TempDir::new().unwrap();
let scope = Scope::writable(dir.path()).unwrap();
let session: WorkdirSessionHandle =
Arc::new(LocalWorkdirSession::new(scope, dir.path().to_path_buf()));
let (_meta, tool) = view_image_tool(session)();
let fixtures: [(&str, &[u8], &str); 4] = [
(
"valid.png",
include_bytes!("fixtures/view-image/valid.png"),
"image/png",
),
(
"valid.jpg",
include_bytes!("fixtures/view-image/valid.jpg"),
"image/jpeg",
),
(
"valid.gif",
include_bytes!("fixtures/view-image/valid.gif"),
"image/gif",
),
(
"valid.webp",
include_bytes!("fixtures/view-image/valid.webp"),
"image/webp",
),
];
for (name, fixture, mime) in fixtures {
std::fs::write(dir.path().join(name), fixture).unwrap();
let output = call(&tool, json!({ "path": name })).await;
let agen::tool::Attachment::Image(image) = &output.attachments[0];
assert_eq!(image.mime_type(), mime);
assert_eq!(image.data(), fixture);
}
}
#[tokio::test]
async fn delegated_view_image_preserves_caller_relative_path_identity() {
let dir = TempDir::new().unwrap();
std::fs::create_dir(dir.path().join("docs")).unwrap();
let fixture = include_bytes!("fixtures/view-image/valid.png");
std::fs::write(dir.path().join("docs/image.png"), fixture).unwrap();
let source: WorkdirSessionHandle = Arc::new(LocalWorkdirSession::new(
Scope::writable(dir.path()).unwrap(),
dir.path().to_path_buf(),
));
let broker = WorkdirToolBroker::new(source);
let lease = broker
.scope(WorkdirToolScope {
rules: vec![WorkdirToolScopeRule {
target: WorkdirPath::new("docs").unwrap(),
permission: WorkdirToolScopePermission::Read,
recursive: true,
symlink_policy: Default::default(),
}],
cwd: WorkdirPath::new("docs").unwrap(),
command: false,
})
.await
.unwrap();
let (_meta, tool) = view_image_tool(lease.tool_session())();
let output = call(&tool, json!({ "path": "image.png" })).await;
assert!(output.summary.contains("image.png"));
assert!(!output.summary.contains("docs/image.png"));
let agen::tool::Attachment::Image(image) = &output.attachments[0];
assert_eq!(image.data(), fixture);
lease.close().await.unwrap();
broker.close().await.unwrap();
}
#[tokio::test]
async fn view_image_reads_scoped_bytes_into_durable_tool_detail() {
let dir = TempDir::new().unwrap();
let scope = Scope::writable(dir.path()).unwrap();
let session: WorkdirSessionHandle =
Arc::new(LocalWorkdirSession::new(scope, dir.path().to_path_buf()));
let png = include_bytes!("fixtures/view-image/valid.png");
std::fs::write(dir.path().join("image.png"), png).unwrap();
let definition = view_image_tool(session);
let (_meta, tool) = definition();
let output = call(&tool, json!({ "path": "image.png" })).await;
assert_eq!(output.attachments.len(), 1);
let agen::tool::Attachment::Image(image) = &output.attachments[0];
assert_eq!(image.mime_type(), "image/png");
assert_eq!(image.data(), png);
let serialized = serde_json::to_string(&output).unwrap();
assert!(serialized.contains("attachments"));
let restored: agen::tool::ToolOutput = serde_json::from_str(&serialized).unwrap();
let agen::tool::Attachment::Image(restored_image) = &restored.attachments[0];
assert_eq!(restored_image.data(), png);
let escaped = call_err(&tool, json!({ "path": "../outside.png" })).await;
assert!(escaped.to_string().contains("scope") || escaped.to_string().contains("path"));
}
#[tokio::test]
async fn read_then_edit_then_read_roundtrip() {
let (dir, reg) = setup();
let file = dir.path().join("a.txt");
std::fs::write(&file, "hello world\n").unwrap();
let p = "a.txt";
let read = reg.get("Read");
let edit = reg.get("Edit");
// Read
let r = call(&read, json!({ "file_path": p })).await;
assert!(r.content.unwrap().contains("hello world"));
// Edit (unique replacement)
let e = call(
&edit,
json!({
"file_path": p,
"old_string": "world",
"new_string": "universe",
}),
)
.await;
assert!(e.summary.contains("1 replacement"));
assert_eq!(std::fs::read_to_string(&file).unwrap(), "hello universe\n");
// Re-read reflects the change
let r2 = call(&read, json!({ "file_path": p })).await;
assert!(r2.content.unwrap().contains("hello universe"));
}
#[tokio::test]
async fn write_then_grep_finds_content() {
let (dir, reg) = setup();
let write = reg.get("Write");
let grep = reg.get("Grep");
let file = dir.path().join("notes.txt");
call(
&write,
json!({
"file_path": file.file_name().unwrap().to_str().unwrap(),
"content": "alpha\nNEEDLE\nomega\n",
}),
)
.await;
let g = call(
&grep,
json!({
"pattern": "NEEDLE",
"output_mode": "content",
}),
)
.await;
let body = g.content.unwrap();
assert!(body.contains("notes.txt"));
assert!(body.contains("NEEDLE"));
}
#[tokio::test]
async fn glob_finds_written_files() {
let (_dir, reg) = setup();
let write = reg.get("Write");
let glob = reg.get("Glob");
for name in ["one.md", "two.md", "three.txt"] {
call(
&write,
json!({
"file_path": name,
"content": "x",
}),
)
.await;
}
let g = call(&glob, json!({ "pattern": "*.md" })).await;
let body = g.content.unwrap();
assert!(body.contains("one.md"));
assert!(body.contains("two.md"));
assert!(!body.contains("three.txt"));
}
#[tokio::test]
async fn absolute_path_is_rejected() {
let (_dir, reg) = setup();
let outside = TempDir::new().unwrap();
let write = reg.get("Write");
let err = call_err(
&write,
json!({
"file_path": outside.path().join("x.txt").to_str().unwrap(),
"content": "x",
}),
)
.await;
// Absolute paths are rejected at the logical WorkdirSession boundary.
let msg = format!("{err}");
assert!(
msg.contains("invalid logical filesystem path"),
"unexpected: {msg}"
);
}
#[tokio::test]
async fn write_to_existing_without_read_fails() {
let (dir, reg) = setup();
let file = dir.path().join("exists.txt");
std::fs::write(&file, "preexisting").unwrap();
let write = reg.get("Write");
let err = call_err(
&write,
json!({
"file_path": file.file_name().unwrap().to_str().unwrap(),
"content": "new",
}),
)
.await;
let msg = format!("{err}");
assert!(msg.contains("has not been read"), "unexpected: {msg}");
}
#[tokio::test]
async fn shared_workdir_across_tools() {
// The key invariant: all builtin tools share the same WorkdirSession instance,
// so read-history set by Read is visible to Edit and Write.
let (dir, reg) = setup();
let file = dir.path().join("shared.txt");
std::fs::write(&file, "one\n").unwrap();
let read = reg.get("Read");
let write = reg.get("Write");
// Read via Read tool
call(
&read,
json!({ "file_path": file.file_name().unwrap().to_str().unwrap() }),
)
.await;
// Write via Write tool — must succeed because the shared WorkdirSession has the read
call(
&write,
json!({
"file_path": file.file_name().unwrap().to_str().unwrap(),
"content": "two\n",
}),
)
.await;
assert_eq!(std::fs::read_to_string(&file).unwrap(), "two\n");
}
#[tokio::test]
async fn edit_requires_read_across_tools() {
let (dir, reg) = setup();
let file = dir.path().join("a.txt");
std::fs::write(&file, "foo\n").unwrap();
let edit = reg.get("Edit");
// No prior Read — Edit should fail
let err = call_err(
&edit,
json!({
"file_path": file.file_name().unwrap().to_str().unwrap(),
"old_string": "foo",
"new_string": "bar",
}),
)
.await;
let msg = format!("{err}");
assert!(msg.contains("has not been read"), "unexpected: {msg}");
}
#[tokio::test]
async fn deterministic_tool_order_is_registration_order() {
let (_dir, reg) = setup();
// Registration order from core_builtin_tools(): Read, Write, Edit, Glob, Grep, Bash
let names: Vec<&str> = reg.entries.iter().map(|(m, _)| m.name.as_str()).collect();
assert_eq!(
names,
vec!["Read", "Write", "Edit", "Glob", "Grep", "Bash",]
);
}
// Regression: tool name capitalization matches Claude Code reference
#[test]
fn tool_names_match_reference_spec() {
let (_dir, reg) = setup();
for expected in ["Read", "Write", "Edit", "Glob", "Grep", "Bash"] {
assert!(
reg.entries.iter().any(|(m, _)| m.name == expected),
"missing tool {expected}"
);
}
}
#[tokio::test]
async fn tracker_recent_files_tracks_read_write_edit() {
// Build a fresh registry that shares a tracker we can query afterwards.
let dir = TempDir::new().unwrap();
let scope = Scope::writable(dir.path()).unwrap();
let fs: WorkdirSessionHandle =
Arc::new(LocalWorkdirSession::new(scope, dir.path().to_path_buf()));
let tracker = Tracker::new();
let reg = Registry::new(core_builtin_tools(
fs,
tracker.clone(),
tools::bash_output::BashOutputView::new(),
));
let a = dir.path().join("a.txt");
let _b = dir.path().join("b.txt");
std::fs::write(&a, "one\n").unwrap();
// Read `a` — should appear in recency.
call(&reg.get("Read"), json!({ "file_path": "a.txt" })).await;
// Write `b` (new file) — should appear ahead of `a`.
call(
&reg.get("Write"),
json!({ "file_path": "b.txt", "content": "hello\n" }),
)
.await;
// Edit `a` — should bump it back to the front.
call(
&reg.get("Edit"),
json!({
"file_path": "a.txt",
"old_string": "one",
"new_string": "two",
}),
)
.await;
let recent = tracker.recent_files(10);
assert_eq!(recent.len(), 2);
assert!(
recent[0].ends_with("a.txt"),
"front should be a.txt: {recent:?}"
);
assert!(
recent[1].ends_with("b.txt"),
"second should be b.txt: {recent:?}"
);
}
#[tokio::test]
async fn bash_inherits_workdir_cwd() {
// The Bash tool starts at the WorkdirSession's pwd. Without any `cd`, its
// `pwd` should canonicalize to the workspace root we set up.
let (dir, reg) = setup();
let bash = reg.get("Bash");
let out = call(&bash, json!({ "command": "pwd", "cwd": "." })).await;
let body = out.content.unwrap();
let parsed: serde_json::Value = serde_json::from_str(&body).unwrap();
let actual = std::fs::canonicalize(parsed["stdout_preview"].as_str().unwrap().trim()).unwrap();
let expected = std::fs::canonicalize(dir.path()).unwrap();
assert_eq!(actual, expected);
}
fn output_fixture() -> (
TempDir,
WorkdirSessionHandle,
tools::bash_output::BashOutputView,
Arc<dyn Tool>,
) {
let root = TempDir::new().unwrap();
let session: WorkdirSessionHandle = Arc::new(LocalWorkdirSession::new(
Scope::writable(root.path()).unwrap(),
root.path().into(),
));
let view = tools::bash_output::BashOutputView::new();
let (_, bash) = tools::bash_tool(session.clone(), view.clone())();
(root, session, view, bash)
}
#[tokio::test]
async fn bash_output_is_retained_without_a_filesystem_grant_or_telemetry_subscriber() {
let (root, _session, view, bash) = output_fixture();
let out = call(
&bash,
json!({ "command": "printf 'x%.0s' {1..20480}; printf '\\377€' >&2; exit 7", "cwd": "." }),
)
.await;
let body: serde_json::Value = serde_json::from_str(&out.content.unwrap()).unwrap();
assert_eq!(body["output"]["terminal"]["exit_code"], 7);
assert_eq!(body["preview_omitted"], true);
assert_eq!(body["output"]["storage_truncated"], false);
let id = body["output"]["id"].as_str().unwrap();
assert_eq!(
view.read(id, workdir::CommandStream::Stdout, 0, 65536)
.unwrap(),
vec![b'x'; 20480]
);
assert_eq!(
view.read(id, workdir::CommandStream::Stderr, 0, 65536)
.unwrap(),
[vec![255], "€".as_bytes().to_vec()].concat()
);
assert!(
body["stdout_entry"]
.as_str()
.unwrap()
.starts_with("/bash-output/")
);
assert!(!body.to_string().contains(root.path().to_str().unwrap()));
assert_eq!(std::fs::read_dir(root.path()).unwrap().count(), 0);
}
async fn await_command_output(
events: &mut tokio::sync::broadcast::Receiver<workdir::CommandEvent>,
) {
tokio::time::timeout(std::time::Duration::from_secs(5), async {
loop {
if matches!(
events.recv().await.unwrap(),
workdir::CommandEvent::Output { .. }
) {
break;
}
}
})
.await
.expect("command output readiness");
}
async fn await_command_terminal(
events: &mut tokio::sync::broadcast::Receiver<workdir::CommandEvent>,
) {
tokio::time::timeout(std::time::Duration::from_secs(5), async {
loop {
if let workdir::CommandEvent::Terminal { status, .. } = events.recv().await.unwrap() {
assert_eq!(status, workdir::CommandStatus::Cancelled);
break;
}
}
})
.await
.expect("command terminal fence");
}
#[tokio::test]
async fn bash_cancellation_returns_bounded_progress_as_terminal_output() {
let (_root, session, _view, bash) = output_fixture();
let mut events = session.subscribe_command_events().unwrap();
let context = ToolExecutionContext::new("call-heavy", "attempt-heavy", 0);
let executing = bash.clone();
let execution_context = context.clone();
let execution = tokio::spawn(async move {
executing.execute(r#"{"command":"printf 'before'; printf 'err-before' >&2; while :; do :; done","cwd":"."}"#, execution_context).await
});
await_command_output(&mut events).await;
bash.cancel_execution(&context).await.unwrap();
let ToolError::Cancelled(output) =
tokio::time::timeout(std::time::Duration::from_secs(5), execution)
.await
.unwrap()
.unwrap()
.unwrap_err()
else {
panic!("expected cancellation");
};
let body: serde_json::Value = serde_json::from_str(&output.content.unwrap()).unwrap();
assert_eq!(body["stdout_preview"], "before");
assert_eq!(body["stderr_preview"], "err-before");
assert_eq!(body["output"]["terminal"]["status"], "cancelled");
assert_eq!(body["output"]["terminal"]["transport_complete"], true);
}
#[tokio::test]
async fn bash_force_close_cleanup_stops_command_and_keeps_session_reusable() {
let (_root, session, _view, bash) = output_fixture();
let mut events = session.subscribe_command_events().unwrap();
let context = ToolExecutionContext::new("call-force", "attempt-force", 0);
let (handle, terminal) = ToolExecutionHandle::start(
bash.clone(),
r#"{"command":"printf ready; while :; do :; done","cwd":"."}"#.into(),
context,
);
await_command_output(&mut events).await;
handle.force_close();
assert!(matches!(
terminal.await,
ToolExecutionTerminal::OutcomeUnknown
));
await_command_terminal(&mut events).await;
let output = call(&bash, json!({"command":"printf reused", "cwd":"."})).await;
let body: serde_json::Value = serde_json::from_str(&output.content.unwrap()).unwrap();
assert_eq!(body["stdout_preview"], "reused");
}
// Sanity: unused Path import guard
const _: fn() -> &'static Path = || Path::new("/");
#[tokio::test]
async fn parent_and_sibling_broker_commands_keep_output_ownership_after_revoke() {
let (root, source, parent_view, parent_bash) = output_fixture();
let broker = WorkdirToolBroker::new(source);
let mut leases = Vec::new();
let mut views = Vec::new();
let mut commands = Vec::new();
for name in ["left", "right"] {
std::fs::create_dir(root.path().join(name)).unwrap();
let lease = Arc::new(
broker
.scope(WorkdirToolScope {
rules: vec![WorkdirToolScopeRule {
target: WorkdirPath::new(name).unwrap(),
permission: WorkdirToolScopePermission::Write,
recursive: true,
symlink_policy: Default::default(),
}],
cwd: WorkdirPath::new(name).unwrap(),
command: true,
})
.await
.unwrap(),
);
let authority = lease.clone();
let view = parent_view.child(Arc::new(move || authority.is_active()));
let (_, bash) = tools::bash_tool(lease.tool_session(), view.clone())();
leases.push(lease);
views.push(view);
commands.push(bash);
}
let left_input =
json!({"command":"printf 'left-%.0s' {1..10000}; printf left-err >&2", "cwd":"."});
let right_input =
json!({"command":"printf 'right-%.0s' {1..10000}; printf right-err >&2", "cwd":"."});
let parent_input = json!({"command":"printf parent; printf parent-err >&2", "cwd":"."});
let (left, right, parent) = tokio::join!(
call(&commands[0], left_input),
call(&commands[1], right_input),
call(&parent_bash, parent_input)
);
let outputs = [left, right, parent]
.map(|out| serde_json::from_str::<serde_json::Value>(&out.content.unwrap()).unwrap());
for (index, label) in ["left", "right", "parent"].iter().enumerate() {
let id = outputs[index]["output"]["id"].as_str().unwrap();
let bytes = parent_view
.read(id, workdir::CommandStream::Stderr, 0, 65536)
.unwrap();
assert_eq!(bytes, format!("{label}-err").as_bytes());
for (child_index, view) in views.iter().enumerate() {
assert_eq!(view.metadata(id).is_some(), child_index == index);
}
}
let left_id = outputs[0]["output"]["id"].as_str().unwrap();
leases[0].close().await.unwrap();
assert!(views[0].metadata(left_id).is_none());
assert!(parent_view.metadata(left_id).is_some());
assert!(
views[1]
.metadata(outputs[1]["output"]["id"].as_str().unwrap())
.is_some()
);
leases[1].close().await.unwrap();
broker.close().await.unwrap();
parent_view.revoke();
assert!(parent_view.metadata(left_id).is_none());
}
#[tokio::test]
async fn utf8_quota_boundary_preserves_raw_prefix_and_reports_saved_vs_seen() {
let (_root, _session, view, bash) = output_fixture();
let out = call(
&bash,
json!({"command":"printf '%4194303s' ''; printf '€'", "cwd":"."}),
)
.await;
let body: serde_json::Value = serde_json::from_str(&out.content.unwrap()).unwrap();
let id = body["output"]["id"].as_str().unwrap();
let saved = body["output"]["stdout_saved"].as_u64().unwrap() as usize;
assert_eq!(saved, 4 * 1024 * 1024);
assert_eq!(body["output"]["stdout_seen"], 4 * 1024 * 1024 + 2);
assert_eq!(body["output"]["stderr_seen"], 0);
assert_eq!(body["output"]["storage_truncated"], true);
assert_eq!(
view.read(id, workdir::CommandStream::Stdout, saved - 1, 10)
.unwrap(),
vec![0xe2]
);
assert!(
view.line_window(id, workdir::CommandStream::Stdout, 0, 1)
.unwrap()
.is_err()
);
assert!(out.summary.starts_with("$ "));
}
#[tokio::test]
async fn timeout_terminal_fences_retained_stdout_and_stderr() {
let (_root, _session, view, bash) = output_fixture();
let error = tokio::time::timeout(std::time::Duration::from_secs(5), bash.execute(r#"{"command":"printf before; printf err-before >&2; while :; do :; done","cwd":".","timeout":1}"#, Default::default())).await.unwrap().unwrap_err();
let ToolError::Interrupted(out) = error else {
panic!("expected timeout output");
};
let body: serde_json::Value = serde_json::from_str(&out.content.unwrap()).unwrap();
let id = body["output"]["id"].as_str().unwrap();
assert_eq!(body["output"]["terminal"]["status"], "timed_out");
assert_eq!(body["output"]["terminal"]["transport_complete"], true);
assert_eq!(
view.read(id, workdir::CommandStream::Stdout, 0, 65536)
.unwrap(),
b"before"
);
assert_eq!(
view.read(id, workdir::CommandStream::Stderr, 0, 65536)
.unwrap(),
b"err-before"
);
}
// Integration: real process pipes and provider replay into the production Bash collector.
// Only response delivery is fault-injected; Runtime's HTTP suite covers actual wire routing.
#[tokio::test]
async fn lost_page_and_terminal_responses_replay_without_restarting_or_double_appending() {
for lose_terminal_only in [false, true] {
let dir = TempDir::new().unwrap();
let session = Arc::new(ExternalValidatingSession {
inner: LocalWorkdirSession::new(
Scope::writable(dir.path()).unwrap(),
dir.path().into(),
),
truncate_first_binary_response: false,
mutate_after_first_binary_response: None,
command_response_losses: std::sync::atomic::AtomicUsize::new(1),
lose_terminal_only,
});
let view = tools::bash_output::BashOutputView::new();
let definitions = core_builtin_tools(session.clone(), Tracker::new(), view.clone());
let bash = Registry::new(definitions).get("Bash");
let out = call(&bash, json!({"cwd":".","command":"printf once >> effect.txt; printf 'x%.0s' {1..200000}; printf err >&2; exit 3"})).await;
let body: serde_json::Value = serde_json::from_str(&out.content.unwrap()).unwrap();
assert_eq!(
std::fs::read(dir.path().join("effect.txt")).unwrap(),
b"once"
);
assert_eq!(
session
.command_response_losses
.load(std::sync::atomic::Ordering::Acquire),
0
);
assert_eq!(body["output"]["stdout_saved"], 200000);
assert_eq!(body["output"]["stdout_delivered"], 200000);
assert_eq!(body["output"]["stderr_saved"], 3);
assert_eq!(body["output"]["terminal"]["exit_code"], 3);
assert_eq!(body["output"]["terminal"]["transport_complete"], true);
assert_eq!(body["output"]["storage_truncated"], false);
let id = body["output"]["id"].as_str().unwrap();
for offset in (0..200000).step_by(65536) {
assert_eq!(
view.read(id, workdir::CommandStream::Stdout, offset, 65536)
.unwrap(),
vec![b'x'; (200000 - offset).min(65536)]
);
}
use workdir::WorkdirSession as _;
session.close().await.unwrap();
}
}