Unified management of the ai-context/ memory layer. Three modes: init, update, maintain.
Triggers: /memory-init, /memory-update, /memory-maintain, initialize memory, update memory, maintain memory, memory housekeeping, clean ai-context
Claude execution wiring
These notes bind the tool-neutral logic above to the Claude Code harness. They are the EXECUTION layer that does not port to other tools.
- Project instruction file — in the Claude harness the neutral "project
instruction file" (maintain mode, gap detection) is
CLAUDE.mdat the project root. - Feature file format — the neutral "feature-domain-expert reference skill"
resolves to
~/.claude/skills/feature-domain-expert/SKILL.md. - Decay review (
mem_review) — maintain mode only, manual, no hook: callmem_review(action: "list")to fetch observations past theirreview_after; for each still-valid row callmem_review(action: "mark_reviewed", observation_id: <id>)to reset decay; stale rows getmem_updateor are left to decay. - Pinning (
mem_pin) — maintain mode pins ONLY canonical topic-key heads of typesarchitecture/decision(policy:_shared/engram-convention.md, "Pinning Policy"). Pins are DEVICE-LOCAL, not synced — never claim a cross-device guarantee.mem_unpinsuperseded heads.