ticket: use base32 project record ids

2026-06-09 22:10:47 +09:00
parent 0803bc3725
commit 4203988d74
798 changed files with 477 additions and 105 deletions
@@ -0,0 +1,32 @@
+Implemented and merged conditional memory prompt guidance.
+
+Merged commits:
+
+- `03256db memory: gate prompt guidance`
+- `3a0c8e1 memory: clean prompt helper warning`
+- `bdb52b1 merge: memory prompt guidance`
+
+Review:
+
+- Review approved in `62d88d9 review: approve memory prompt guidance`.
+- Initial review requested one warning cleanup; follow-up review approved the fix.
+
+Summary:
+
+- Normal prompt memory/knowledge guidance is now gated by available tool capabilities.
+- Memory-disabled prompts do not advertise `MemoryQuery`, `KnowledgeQuery`, `MemoryRead`, or memory mutation tools.
+- Partial-capability prompts avoid naming unavailable memory vs knowledge tools.
+- Normal Pod guidance now encourages somewhat more proactive small targeted lookups when prior decisions, rationale, durable preferences, tickets, policy, workflow, or project history may matter.
+- Existing cautions remain: memory can be stale; current authoritative state lives in user instructions, repository files, tickets, git history, and session logs; memory should not be queried mechanically every turn; mutation tools remain restricted.
+- Internal memory extraction/consolidation prompts remain focused on their worker roles and do not inherit normal Pod guidance.
+
+Validation after merge:
+
+- `cargo fmt --check` — passed
+- `cargo test -p pod prompt::` — passed (`53 passed; 0 failed`)
+- `./tickets.sh doctor` — passed
+- `git diff --check` — passed
+
+Caveats:
+
+- The test run still reports an existing unrelated `llm-worker` dead-code warning; the new prompt helper warning found during review was removed.
@@ -0,0 +1,27 @@
+# Review: memory prompt conditional lookup
+
+Implementation reviewed on branch `memory-prompt-conditional-lookup`.
+
+Reviewed commits:
+
+- `03256db memory: gate prompt guidance`
+- `3a0c8e1 memory: clean prompt helper warning`
+
+Verdict: approve.
+
+Summary:
+
+- Prompt rendering now gates normal memory/knowledge guidance according to available tool capabilities.
+- Memory-disabled prompt rendering does not advertise memory tools.
+- Partial-capability rendering avoids naming unavailable memory or knowledge tools.
+- Normal Pod guidance now encourages somewhat more proactive small targeted lookup when prior decisions, rationale, durable preferences, tickets, policy, workflow, or project history may affect the response or action.
+- Existing cautions remain: memory can be stale; current authoritative state lives in user instructions, repository files, tickets, git history, and session logs; memory should not be queried mechanically every turn; mutation tools are restricted.
+- Internal memory extraction/consolidation worker prompts remain isolated from normal Pod guidance.
+
+Review notes:
+
+- Initial review requested one blocker fix because `cargo test -p pod prompt::` emitted a new unused `append_trailing_section` warning.
+- Follow-up commit `3a0c8e1` removed the warning-producing helper path and adjusted partial-capability wording.
+- Final focused review approved the follow-up. `cargo test -p pod prompt::` passed with `53 passed; 0 failed`; only an existing unrelated `llm-worker` dead-code warning remains.
+
+Merge readiness: ready to merge.
@@ -0,0 +1,67 @@
+---
+title: "Memory prompt: conditional guidance and proactive lookup"
+state: "closed"
+created_at: "2026-05-31T22:35:06Z"
+updated_at: "2026-05-31T22:52:35Z"
+---
+
+## Background
+
+The normal tool-usage prompt currently contains memory guidance unconditionally:
+
+- use targeted lookup for past decisions, prior requests, durable preferences, project history, or rationale;
+- use `MemoryQuery`, `KnowledgeQuery`, and `MemoryRead` appropriately;
+- treat resident memory/knowledge as helpful but stale;
+- avoid querying memory every turn;
+- avoid mutation unless explicitly requested or running a memory maintenance worker.
+
+This is directionally correct, but two prompt-level issues remain:
+
+1. Memory/Knowledge guidance should be rendered only when the corresponding memory capability is enabled and available. If memory is disabled for a Pod/profile/internal worker, the prompt should not advertise tools or behavior that cannot be used.
+2. The read-side guidance is too passive. Agents should be encouraged to use small targeted queries more often when prior project decisions, user preferences, historical rationale, recent ticket work, or policy/workflow context may affect the answer or action.
+
+This should be implemented as prompt/template behavior, not by weakening tool permissions or making memory lookup automatic.
+
+## Requirements
+
+- Gate memory/knowledge prompt text through the prompt/template engine based on whether memory/knowledge tools or resident memory features are enabled for the current worker.
+- Do not show instructions for unavailable memory tools.
+- Keep internal/disposable worker behavior conservative; do not accidentally inject normal Pod memory guidance into memory extraction/consolidation workers or other internal workers that should not see it.
+- Strengthen normal Pod read-side guidance so agents perform small targeted `MemoryQuery` / `KnowledgeQuery` lookups more proactively when history may matter.
+- Preserve the existing warnings:
+  - memory may be stale;
+  - authoritative current state is in user instructions, repository files, tickets, git history, and session logs;
+  - do not query memory mechanically every turn;
+  - mutation tools are for explicit user requests or memory maintenance workers.
+- Do not introduce automatic pre-request memory queries in this ticket.
+- Do not change memory extraction/consolidation semantics except as needed to keep prompt rendering coherent.
+
+## Suggested prompt direction
+
+The final wording does not need to match this exactly, but should convey:
+
+```text
+Use memory and knowledge proactively when the request may depend on prior
+project decisions, historical rationale, durable user preferences, recently
+completed tickets, or established workflow/policy conventions. Prefer a small
+targeted MemoryQuery/KnowledgeQuery before relying on vague recollection.
+
+Strong lookup triggers include: the user says "recently", "previously", "that
+decision", "the ticket", "why", "policy", or "workflow"; you are about to make
+a design recommendation; you are reviewing, merging, closing, or rescoping a
+work item; or you are about to assert project history from memory.
+
+Do not query memory mechanically on every turn. Skip memory lookup for purely
+local facts answered by current repository files, command output, or current
+user instructions.
+```
+
+## Acceptance criteria
+
+- Normal Pod prompts include memory/knowledge guidance only when the corresponding capability is enabled and available.
+- Prompts for memory-disabled profiles/workers do not advertise `MemoryQuery`, `KnowledgeQuery`, `MemoryRead`, or memory mutation tools.
+- Normal Pod memory guidance explicitly encourages somewhat more frequent small targeted queries when history, rationale, preferences, tickets, policy, workflow, or prior decisions may matter.
+- Existing stale/authority cautions and no-mechanical-query guidance remain.
+- Internal memory extraction/consolidation prompts remain focused on their worker roles and do not inherit normal Pod guidance accidentally.
+- Tests or focused snapshots cover enabled vs disabled prompt rendering, or the implementation otherwise includes a clear focused regression test for conditional rendering.
+- `cargo fmt --check`, relevant prompt/worker tests, `./tickets.sh doctor`, and `git diff --check` pass.
@@ -0,0 +1,42 @@
+<!-- event: create author: tickets.sh at: 2026-05-31T22:35:06Z -->
+
+## Created
+
+Created by tickets.sh create.
+
+---
+
+<!-- event: review author: hare at: 2026-05-31T22:51:36Z status: approve -->
+
+## Review: approve
+
+# Review: memory prompt conditional lookup
+
+Implementation reviewed on branch `memory-prompt-conditional-lookup`.
+
+Reviewed commits:
+
+- `03256db memory: gate prompt guidance`
+- `3a0c8e1 memory: clean prompt helper warning`
+
+Verdict: approve.
+
+Summary:
+
+- Prompt rendering now gates normal memory/knowledge guidance according to available tool capabilities.
+- Memory-disabled prompt rendering does not advertise memory tools.
+- Partial-capability rendering avoids naming unavailable memory or knowledge tools.
+- Normal Pod guidance now encourages somewhat more proactive small targeted lookup when prior decisions, rationale, durable preferences, tickets, policy, workflow, or project history may affect the response or action.
+- Existing cautions remain: memory can be stale; current authoritative state lives in user instructions, repository files, tickets, git history, and session logs; memory should not be queried mechanically every turn; mutation tools are restricted.
+- Internal memory extraction/consolidation worker prompts remain isolated from normal Pod guidance.
+
+Review notes:
+
+- Initial review requested one blocker fix because `cargo test -p pod prompt::` emitted a new unused `append_trailing_section` warning.
+- Follow-up commit `3a0c8e1` removed the warning-producing helper path and adjusted partial-capability wording.
+- Final focused review approved the follow-up. `cargo test -p pod prompt::` passed with `53 passed; 0 failed`; only an existing unrelated `llm-worker` dead-code warning remains.
+
+Merge readiness: ready to merge.
+
+
+---