after-action

name: after-action version: 1.0.0 description: | Task wrap-up and self-evolution. Review what happened, extract lessons, archive logs, back up with git. The closing ritual for every work session. Triggers: "wrap up", "archive", "AAR", "after action", "收官" tags: [meta, evolution, memory, archive]

🔄 After Action — Task Wrap-Up & Learning

Done but not archived = not done. Every action feeds the next one.

Core attitude: always find N→N+1. Wrapping up isn't just "recording what happened" — it's "finding what can be better next time." Perfection isn't the goal. Continuous evolution is. Today's system is always tomorrow's starting point.

Triggers

Any of these activate this skill:

• "wrap up" / "archive" / "AAR" / "after action"
• "收官" / "歸檔" / "整合經驗"
• End of a long work session
• After completing a complex multi-step task

🎯 Execution Flow

Step 1: Inventory — What happened?

List everything:

• Tasks completed
• Decisions made
• Files created or modified
• Problems encountered and solved

Step 2: Extract — What did we learn?

From each completed task, pull out:

• Problems → solutions → how to prevent next time
• New best practices discovered
• Performance data (time spent, what worked, what didn't)

Write to daily log: memory/YYYY-MM-DD.md

Step 3: Archive — Write it down

1. Daily log — memory/YYYY-MM-DD.md

• Complete record of the day
• Format: time, event, decision, output, lesson

2. Long-term memory (for significant events) — MEMORY.md

• New cognitive breakthroughs
• Important decisions with context
• Systemic changes (before/after)

Step 4: Reflect — What really happened?

Not mechanical archiving — actually think about it.

Four required questions:

1. What went right — Most valuable output today? Why?
2. What to watch — Any bad patterns forming? Time/energy spent wisely?
3. The real issue — What's the actual underlying thing going on?
4. N→N+1 — Based on everything today, what's the most natural next step? Not "what should I do" but "if I follow today's momentum, where does it lead?"

Tone: Direct, honest, but caring. This isn't a performance review — it's a debrief between partners.

Step 5: Evolve — Grow from experience

1. Lessons → write to relevant skill files or SOUL.md
2. Soul-level updates — If a lesson is systemic, recurring, or violates core principles → must be recorded in SOUL.md
3. Micro-improvements — Spot automatable repetitive tasks → build script/alias
4. N→N+1 list — For each skill/system/process, write one specific "can be better next time" point

Step 6: Backup — Git commit + push

cd /path/to/workspace
git add -A
git commit -m "🔄 after-action: $(date +%Y-%m-%d\ %H:%M) — wrap up"
git push

📋 Verification Checklist (Gate Check)

Rule: Before announcing "wrap-up complete", every item must be checked. Missing any = not complete.

□ Step 1: Inventory — all tasks listed?
□ Step 2: Extract — lessons written?
□ Step 3: Archive — daily log complete?
□ Step 4: Reflect — four questions answered? (not one-liners)
□ Step 5: Evolve — lessons distributed? Soul updated if needed?
□ Step 6: Git — committed and pushed?

Pre-Report Self-Check

Before saying "wrap-up done", ask yourself:

1. Are all boxes checked above?
2. Is there anything I "thought I did" but actually just thought about?
3. If my human asks "what did you do in Step X?", can I give a specific answer?

If any answer is no → go back and finish. Don't announce completion.

⚡ One-Trigger Mode

When your human says "wrap up" or "archive", don't ask questions. Run all steps.

Report format:

🔄 Wrap-up complete

📊 Today's activity:
  - [summary of what was done]

💭 Reflection:
  [What went right / What to watch / The real issue — direct and honest]

🧬 Evolution:
  - X lessons extracted
  - [any soul-level updates]

🔒 Git:
  - ✅ committed: [hash]

🧠 When to auto-trigger?

⚠️ Notes

• Don't build new features during wrap-up — this is organizing, not developing
• Lessons must be specific — "had a problem" isn't enough. What problem? What solution? How to prevent?
• Git message uses 🔄 — distinguishes wrap-up commits from feature commits

"Did it but didn't record it = didn't do it. Recorded but didn't push = didn't record."