Agent Task Plan Template
Create a task folder and copy this file to:
tmp/agent-plans/<YYYY-MM-DD>-<task-slug>/plan.md
before implementation.
Recommended per-task structure:
tmp/agent-plans/
<YYYY-MM-DD>-<task-slug>/
plan.md # required source of truth for the task
worklog.md # optional detailed command/output notes
review.md # optional reviewer findings and sign-off
0. Kickoff Message (Send Before Implementation)
Use this 5-line kickoff format in chat before substantial implementation/debugging work:
- Scope:
- Target tests:
- Open questions status:
- Risks:
- Plan:
Open questions status should normally be None (resolved).
If not None (resolved), implementation is blocked until questions are resolved with the human.
1. Task Summary
- Request:
- In scope:
- Out of scope:
2. Clarifications and Decisions
- Open questions:
- Open questions resolved before implementation
- OpenSpec decision:
- Use OpenSpec
- OpenSpec overridden by human ("no OpenSpec")
- If overridden, rationale:
- Options presented to human:
- Option A:
- Option B:
- Option C:
- Human decision:
- Decisions log:
3. Acceptance Targets
Target Tests (Primary Acceptance)
- Use
docs/dev/component_command_registry.mdto select canonical commands. - Test command 1 (stack-appropriate):
- Test command 2 (stack-appropriate):
- New/updated tests to add:
Non-Test Acceptance Criteria
- Criterion 1:
- Criterion 2:
4. Edge Cases and Risks
- Edge case 1:
- Edge case 2:
- Risk and mitigation:
5. Test-First Execution Log
- Wrote/updated tests before any production code changes
- Verified failing tests before code changes
- Recorded failing test command/result before first production code change
- Implemented production code changes
- Re-ran target tests until green
- Ran regression/adjacent tests
- No process violation (no production code edits before failing-test record)
5a. Process Violations (If Any)
- Violation occurred? (yes/no):
- What happened:
- Corrective action taken:
5b. Reviewer Coverage Audit
- Unit-test coverage reviewed
- Integration-style test coverage reviewed (or N/A with reason)
- E2E coverage reviewed via code-path inspection/CI evidence if supported by existing harness (or N/A with reason)
- Confirmed no local E2E run was performed unless explicitly requested by human
- Coverage gaps and residual risk documented
5c. Reviewer Output (Required Format)
- Findings by priority:
P0:P1:P2:P3:- Use
docs/dev/review_severity_rubric.mdfor priority assignment. - Coverage audit summary:
- Unit:
- Integration-style:
- E2E:
- Residual risk summary:
- Final recommendation:
approve/needs changes
6. Workstream Plan (If Parallelized)
- Workstream A:
- Scope:
- Worktree:
- Implementor:
- Reviewer:
- Workstream B:
- Scope:
- Worktree:
- Implementor:
- Reviewer:
7. Merge and Verification Checklist
- Manually inspected each worktree diff
- Merged workstreams one by one
- Ran target tests after each merge
- Committed after each successful merge
8. Progress Log
- Timestamp:
- Change:
- Test result:
- Commit:
9. Process Improvements
- Did human corrections or debugging-loop guidance occur? (yes/no):
- If yes, proposed update location:
AGENTS.mdpath:- or local note in
tmp/agent-learnings/<YYYY-MM-DD>-<topic>.md: - Proposed rule/procedure update: