memory-playbooklisted

# Memory Playbook A protocol for Claude Code's file-based memory that solves two chronic problems: 1. **The compaction death spiral.** A single MEMORY.md that holds running status grows until it hits the read limit, gets compacted, loses nuance, grows again. Fix: the index NEVER holds status — it only holds rules and a map. Growth happens in per-domain files, each loaded only when relevant. 2. **Repeating the same corrections.** The owner corrects the same class of mistake across sessions because lessons die with the conversation. Fix: an explicit capture → store → replay loop (the playbook), so corrections become pre-ship checks instead of recurring complaints. The memory directory is wherever this project's harness keeps it (typically `~/.claude/projects/<project-slug>/memory/`). Write memory files in the language the owner uses with you. ## The architecture (three layers + two special files) ``` memory/ ├── MEMORY.md # INDEX ONLY: usage rules + domain map + cross-domain one-liners. │ # Auto-loaded every session. Never holds status → never bloats. ├── project-map.md # STABLE invariants: what the system IS (business, repos, │ # architecture rules, verification protocol). No status here. ├── playbook.md # SELF-LEARNING: owner's philosophy, red flags that trigger │ # corrections, mandatory pre-ship self-review, lessons journal. ├── domain-<area>.md # One per work area. Holds