memory-md-cleanup

MEMORY.md Cleanup

~/.{{AGENT_NAME_LOWER}}/P2-hippocampus/memories/MEMORY.md hits 8K char cap when auto-accumulated + user entries pile up. Manual cleanup needed — auto-cleanup is NOT implemented (growth-2026.md "분기별 메모리 정리 자동화" is TODO).

Trigger

• System prompt shows [98% — 7,901/8,000 chars] or similar near-cap status
• User says "메모리 정리" / "MEMORY.md 정리"

Steps

1. Read current state

content = open('~/.{{AGENT_NAME_LOWER}}/P2-hippocampus/memories/MEMORY.md').read()
print(f'chars: {len(content)}, cap usage: {len(content)/8000*100:.1f}%, entries (§): {content.count(chr(167))}')

2. Check for concurrent writes

MEMORY.md.lock mtime 5분+ stale이 아니면 wait. 그 외 진행.

3. Classify entries

Cut (resolved/one-time):

• "follow-up" / "patched" / "fixed" / "완료" / "✅" 단어 등장
• 다른 entry가 미참조하는 historical event
• system prompt active docs (SELF 모델, KANBAN_INDEX, architecture-dataflow)에서 미참조

Keep (operational/active):

• system prompt active docs에서 참조되는 facts
• port / path / plist label / version / token / CF account 같은 operasional numbers
• 다음 session에 적용될 trigger pattern (self-critique framing, cron infra, mock patterns)
• ongoing incident 핵심 findings (resolve되기 전)

4. Present options via mcp_clarify

H1: aggressive — cut all candidates (0 risk, max headroom) H2: conservative — cut 1~2 largest만 H3: increase cap (8000→12000, config edit, 매 session inject 비용 증가)

User timeout → best judgement = H1. 0 risk, 가장 많은 buffer, follow-up으로 추가 trim 가능.

5. Write new file with mcp_write_file

• Line 1: [YYYY-MM-DD cleanup: H{N}, removed {N} entries (~{chars} saved). Active entries preserved.]
• Following: original format — entry line + § on its own line as separator

6. Fix stale cross-references

제거된 entry를 참조하는 warning이 다른 entry에 남아있으면 patch로 정리. 예: "⚠️ bot.py의 M2.7 호출은 6/1 follow-up patch로 M3 통일됨" — 제거된 follow-up entry를 가리키던 warning을 resolved 상태로 update.

7. Verify

content = open('~/.{{AGENT_NAME_LOWER}}/P2-hippocampus/memories/MEMORY.md').read()
assert len(content) < 8000, f'still over cap: {len(content)}'
print(f'OK: {len(content)} chars ({len(content)/8000*100:.1f}% of 8K cap)')