菜单
Explains how claude-mem captures observations, when memory injection kicks in, where data lives, and how to verify it's running. TRIGGER when: user asks "how does claude-mem work?", "what is this thing doing?", "is claude-mem running?", "where does it store data?", or "explain the memory system".
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | how-it-works |
| description | Explains how claude-mem captures observations, when memory injection kicks in, where data lives, and how to verify it's running. TRIGGER when: user asks "how does claude-mem work?", "what is this thing doing?", "is claude-mem running?", "where does it store data?", or "explain the memory system". |
| origin | claude-mem |
| owner | surfingalien |
You're a pragmatic executor focused on shipping results and measuring impact. You use AI to amplify your effect and automation to eliminate busywork.
Every Read, Edit, and Bash that Claude makes turns into a compressed observation. Observations are summarized at session end. Relevant ones are auto-injected into future prompts so the next session starts with context from the last one.
No re-explaining the codebase. No re-discovering decisions. No starting cold.
Run /learn-codebase if you want to front-load the entire repo into memory in a single session (~5 minutes, optional but powerful for large repos).
Every tool call generates an observation:
| Tool | What's Captured |
|---|---|
| Read | File path, key content summary |
| Edit / Write | File path, change summary, before/after |
| Bash | Command, output summary, exit code |
| grep / glob | Search query, result count |
Observations are compressed by Claude — not raw tool output. They're semantic summaries, not logs.
Everything stays local:
~/.claude-mem/
├── claude-mem.db # SQLite — all observations
├── chroma/ # Vector store — semantic search index
└── sessions/ # Per-session metadata
No cloud sync. No external API calls for storage. Your project history stays on your machine.
# Check worker status
npm run worker:status --prefix ~/.claude/plugins/marketplaces/thedotmack
# Or via npx
npx claude-mem worker:status
# Check observation count
npx claude-mem stats
# Search for recent observations
npx claude-mem search "recent" --limit 5
If the worker isn't running, start it:
npx claude-mem worker:start
At session start (SessionStart hook), the worker:
<memory> block at the start of the session (if verbose mode is on)The injection is additive — it doesn't replace your CLAUDE.md or project context, it prepends it.
claude-mem supports workflow modes via the CLAUDE_MEM_MODE environment variable:
# Standard coding mode (default)
CLAUDE_MEM_MODE=code claude
# Relaxed / chill mode
CLAUDE_MEM_MODE=code--chill claude
# Language-specific (e.g., for code comments in Spanish)
CLAUDE_MEM_MODE=code--es claude
# Email investigation mode (for ragtime)
CLAUDE_MEM_MODE=email-investigation claude
Modes are defined in ~/.claude/plugins/marketplaces/thedotmack/plugin/modes/.
claude-mem v13 ships a Server Beta mode with PostgreSQL + Redis backing:
# Start the server
claude-mem server start
# Check status
claude-mem server status
# Stop
claude-mem server stop
Server Beta adds a /v1 REST API for observations, sessions, memories, and search — useful for integrating claude-mem data into external tools (e.g., FinSurfing dashboard showing your own coding history).
Default install still uses SQLite worker — Server Beta is opt-in.
# Search past observations
npx claude-mem search "anthropic api"
# Generate a summary of recent work
npx claude-mem summary --since "7 days ago"
# Export observations for a project
npx claude-mem export --project /path/to/project --output history.json
# Clean old observations (> 90 days)
npx claude-mem clean --older-than 90d
# Check version
npx claude-mem --version
# Repair if something's broken
npx claude-mem repair
# Install
npx claude-mem install
# Update to latest
npx claude-mem install --update
# Repair broken installation
npx claude-mem repair
mem-search — Search observation historyknowledge-agent — Build queryable brains from observation corporatimeline-report — Narrative project history from observationsweekly-digests — Week-by-week chapter digestBug → spec protocol. When a bug is found or a test fails, trace the cause, decide whether a new §V invariant would catch recurrence, append to §B. This is the one non-obvious thing SDD does that plan-then-execute doesn't. Triggers on test failure, bug report, post-mortem, or explicit user ask.
Plan-then-execute implementation against SPEC.md. Native single-thread loop, no sub-agents. On test or build failure, auto-invokes the backprop skill before retrying — a failed verification always considers whether a new §V invariant would prevent recurrence. Triggers when the user asks to build, implement, execute the spec, or tackle a specific §T task (`build §T.3`, `build --next`, `implement next task`, `run the build`). Expects SPEC.md to exist; if not, defers to the spec skill.
Read-only drift detector. Diffs SPEC.md against current code and reports violations grouped by severity. Writes nothing — suggests remedies via the spec or build skills but never invokes them. Triggers when the user asks to check drift, audit the spec, verify invariants, or ask whether code still matches the spec. Phrasings: "check drift", "audit the spec", "does the code still match §V", "check invariants", "spec vs code".
Create, amend, or backprop bugs into SPEC.md at repo root. Sole mutator of the project spec. Triggers when the user asks to write a spec, start a new spec, distill a spec from existing code, add invariants, amend sections (§G, §C, §I, §V, §T, §B), or record a bug via backprop. Common phrasings: "write the spec for...", "new spec", "bug: ...", "amend §V.3", "distill spec from code", "spec this idea". Reads and follows FORMAT.md for the caveman encoding rules and pipe-table shape of §T and §B.
Decision guide for delegating to caveman-style subagents. Tells the main thread WHEN to spawn `cavecrew-investigator` (locate code), `cavecrew-builder` (1-2 file edit), or `cavecrew-reviewer` (diff review) instead of doing the work inline or using vanilla `Explore`. Subagent output is caveman-compressed so the tool-result injected back into main context is ~60% smaller — main context lasts longer across long sessions. Trigger: "delegate to subagent", "use cavecrew", "spawn investigator/builder/reviewer", "save context", "compressed agent output".
Ultra-compressed commit message generator. Cuts noise from commit messages while preserving intent and reasoning. Conventional Commits format. Subject ≤50 chars, body only when "why" isn't obvious. Use when user says "write a commit", "commit message", "generate commit", "/commit", or invokes /caveman-commit. Auto-triggers when staging changes.