session-retro

End-of-session sync. Mirrors what the offline extraction script does, but for the current conversation.

Steps

1. Count user prompts in this session — scan back through the conversation. Group by date if it spans days.
2. List files modified — anything you Write or Edit'd this session.
3. Identify the headline — what was the main thing accomplished?
4. Find the next session number — list Aqarkom_Knowledge/History/Conversations/Session*.md and pick the next NN.
5. Create Aqarkom_Knowledge/History/Conversations/Session NN - YYYY-MM-DD.md with:
   • Frontmatter (session_id, started, ended, prompts)
   • One-line headline
   • Tool usage summary (which tools, roughly how many calls)
   • Files modified (top 30, deduplicated)
   • User prompts in chronological order with timestamps (best-effort if exact ts not available)
   • Lessons / non-obvious takeaways
   • Related links (Sessions/E*, Decisions, Features)
6. Update Aqarkom_Knowledge/History/Timeline.md — add a row for this session.
7. Update Aqarkom_Knowledge/History/Conversations/Index.md — add this session to the table.
8. Refresh memory — for any guidance the user gave that wasn't already a memory entry, save it now (feedback type).
9. Commit the vault changes with /commit-and-track.

Verification

• New session note exists
• Timeline + Conversations Index updated
• At least one new memory saved IF the user gave new guidance
• Commit lands

Anti-patterns

• Don't dump every prompt verbatim — summarize when prompts are repetitive ("user iterated 5x on growth badge styling")
• Don't include code blobs — link to commits instead
• Don't write the retro for a session where nothing was actually done