| name | promote |
| description | Extract key learnings from daily logs and promote them to identity/memory.md. Run automatically via launchd once per day, or manually after a productive day. |
Daily log promotion — extract signal from raw conversation logs into long-term memory.
Hot/Cold Memory Architecture
- Hot:
identity/memory.md — always in context, kept under 2500 tokens. Only the essentials.
- Cold:
memory/*.md — topic files searched on-demand. Detailed context lives here.
memory/projects.md — project details, architecture, status
memory/preferences.md — recurring patterns, style choices
memory/decisions.md — key decisions with reasoning
memory/people.md — people context
memory/archive-YYYY-MM.md — older entries rotated out of hot memory
Steps
- Read the last 3 days of
daily-logs/ files
- Read current
identity/memory.md (hot) and relevant memory/*.md files (cold)
- Extract and route:
- Key decisions (with rationale) →
identity/memory.md Current State + memory/decisions.md
- Lessons learned →
memory/decisions.md
- New preferences →
memory/preferences.md
- Project context →
memory/projects.md
- People context →
memory/people.md
- Active work / open threads →
identity/memory.md Active Work
- For
identity/memory.md (hot layer):
- Only promote things needed in every conversation
- Update
## Current State and ## Active Work
- Add one-liner to
## Session Log
- Keep under 2500 tokens — push detail to cold files
- For
memory/*.md (cold layer):
- Add detailed entries with datestamps:
<!-- promoted YYYY-MM-DD -->
- Don't duplicate — consolidate with existing entries
- Remove stale entries that are clearly no longer relevant
- If
identity/memory.md is over 2500 tokens, archive oldest entries to memory/archive-YYYY-MM.md
- Report what was promoted, where it went, and what was skipped