claude-md-overhaullisted

# /claude-md-overhaul — Tier-by-tier audit + improvement of the Claude Code memory layer Companion to `/maintain` (which covers MCP/sync/tool-usage health). This skill focuses on the **content + structure** of CLAUDE.md and MEMORY.md. ## When to use - CLAUDE.md feels bloated or stale; specific entries are obviously wrong (skills that don't exist, machines that moved, JIRA-style references after migration) - Old memories aren't surfacing at session start — MEMORY.md is probably past the 25 KB / 200-line auto-load cap and you only see the recent ~30 entries - You added projects but never wrote memory files for them - After a multi-machine migration, you suspect Syncthing has sync-conflict files in `~/.claude/` - Periodic hygiene: roughly monthly, or after a heavy session of new skill/project shipping ## When NOT to use - Anthropic config issues (use `/maintain`) - Pure MCP health (use `/maintain` or `/raven-mcp-doctor`) - Single-fix triage (just edit the file directly) ## Anthropic's documented caps (the targets) | File | Hard cap | Sweet spot | |------|----------|------------| | Any `CLAUDE.md` (user, machine, project) | 200 lines (adherence drops past this) | 80–150 lines | | Project `MEMORY.md` (auto-memory index) | **first 200 lines OR 25 KB loads at session start** — past that is invisible | <150 lines, <20 KB for headroom | Source: [code.claude.com/docs/en/memory](https://code.claude.com/docs/en/memory) ## Procedure Follow tier-by-tier. After each tier, verify c