Project Memory Autopilot

Implement external memory as first-class project infrastructure, not ad-hoc notes.

Workflow

1. Determine runtime target.
   • codex -> .codex/memory/
   • claude -> .claude/memory/
   • opencode -> .opencode/memory/
   • fallback -> .ai/memory/
2. Inspect existing routing/protocol files.
   • Common locations: AGENTS.md, docs/guides/protocol-*.md, local collaboration skill files.
3. Bootstrap memory storage.
   • Prefer script:
     python scripts/bootstrap_memory.py --root <repo> --runtime <codex|claude|opencode|generic>
   • Use --dry-run first, then run without it.
4. Wire memory rules into protocol docs.
   • Read memory files at non-trivial task start.
   • Define hard/soft write triggers.
   • Require final report line:
     Memory Update: written|skipped + files + trigger
5. Seed initial memory entries.
   • User preferences and communication constraints.
   • Active context and next priorities.
   • Decision log entry for the current change.
6. Validate before completion.
   • Memory files exist.
   • Protocol docs reference the memory path and trigger behavior.
   • Final report format includes the memory update line.

Trigger Matrix

Read references/memory-trigger-matrix.md and apply:

• hard trigger: any 1 -> write memory
• soft trigger: any 2 -> write memory

Templates

Use assets as copy-ready templates:

• assets/user-profile.template.md
• assets/active-context.template.md
• assets/decision-log.template.jsonl
• assets/agents-memory-block.template.md
• assets/protocol-memory-block.template.md

Implementation Notes

• Keep memory updates minimal and append-only where possible.
• Store durable preferences in user-profile.md.
• Store short-lived execution state in active-context.md.
• Append key decisions to decision-log.jsonl; do not rewrite history.