memory-maplisted
Install: claude install-skill shandar/memory-map
# Memory Map Skill — v0.2
A two-tier project memory protocol for context that survives long absences.
## The model
Two layers, each with a distinct job:
| Tier | Where it lives | What it stores | Scope |
|------|----------------|----------------|-------|
| **Global** | claude.ai: memory API (`memory_user_edits`) · Claude Code: `~/.claude/memory.md` | Cross-project index + state snapshot | ~30 entries (claude.ai) · under 200 lines (Claude Code) |
| **Per-project** | `<project>/STATUS.md` | Deep state for one project | Unlimited, one per repo |
Global memory answers: *what am I working on across everything?*
STATUS.md answers: *where exactly did I leave off in this one project?*
Read together, they restore full context. Written together, they stay in sync.
---
## Two modes
Identify the mode before doing anything else.
### WRITE mode — save state
Triggered when:
- User asks to save: "save this", "memorise this", "snapshot", "checkpoint [project]"
- User asks to update: "update memory", "store this session", "remember all of this"
- Session naturally closes after productive work (proactive trigger)
- PWP phase transition mentioned ("plan done, moving to execute" → update phase tag)
### READ mode — restore context
Triggered when:
- User asks for orientation about one project: "catch me up on [project]",
"restore context", or "where did I leave off" / "what was I doing" when a project
is named or the cwd has a STATUS.md
- User asks for status: "[project] status", "