wisdom

Wisdom

Post-execute report generation. Captures what was built, deviations applied, patterns discovered, and decisions deferred. Runs automatically at /ac:execute completion; can also run standalone against any status: complete plan.

---

Phase 1: Load

1. Parse $ARGUMENTS:

   • File path → single-plan or Mode B phase wisdom.
   • Directory path → multi-phase Mode A consolidation.
   • Empty → Glob .ac/plans/**/*.md, filter status: complete without a SUMMARY alongside, pick the most recent. If more than one candidate, AskUserQuestion to pick.

2. Read the plan file (or all phase files for Mode A) and any existing SUMMARY / open-questions files.

3. Extract the status: executing timestamp from the frontmatter history or the first commit SHA matching the slug. This is the start of the execution window.

4. Run Bash to collect commit and change data:

   START_REF=<commit or ISO timestamp>
   git log --since="$START_REF" --oneline --no-merges | grep -E "<slug>" || \
     git log --since="$START_REF" --oneline --no-merges
   git diff --stat "$START_REF"..HEAD
   

5. Pre-invoke Skill("my-language") for prose tone. Silent no-op if not installed.

6. Parse the plan file's Deviations Log and Open Questions sections into structured lists for Phase 2 and Phase 3.

---

Phase 2: Generate SUMMARY.md

Target:

• Multi-phase Mode A: .ac/plans/<slug>/SUMMARY.md (append or replace this phase's section).
• Single-file or Mode B: .ac/plans/<slug>-summary.md.

Format:

---
slug: <slug>
completed: <ISO date>
mode: <A | B | single>
phase: <N or "single">
task-count: <N>
deviations: <N>
commits: <N>
---

## What was built

One concise line per completed Task:
- Task 1 (<primary file>): <outcome in one sentence>
- Task 2 (<primary file>): <outcome>

## Key files changed

- `path/to/file.ts`, <what changed, why, one sentence>
- `path/to/other.php`, <what changed>

## Deviations applied

Grouped by rule:
- Rule 1 (bugs): <count>, examples: <one liner per notable case>
- Rule 2 (critical additions): <count>, examples
- Rule 3 (blocking fixes): <count>, examples
- Rule 4 (user asks): <count>, examples

## Patterns discovered

- Existing helper `path/to/x.ts:fnName` matched Task <N> needs, saved an implementation.
- <pattern name>, observed when <context>, reusable for <future scenario>.

## Verification status

- Per-task Verify commands: all passed.
- End-to-end Verification (from plan): <result>.
- Build: <pass | fail>.
- Lint: <zero warnings | N warnings>.

## Open questions forwarded

See `<path>/open-questions.md` or inline (single-file). Count: <N>.