OpenClaw Config Skill 🐾
Manages your openclaw-config installation.
Users say things like:
- "Set up openclaw-config"
- "Update my openclaw config"
- "Force update openclaw" (overwrites their changes)
- "Check if openclaw has updates"
Setup
Do these steps in order:
-
Clone repo to
~/.openclaw-config -
Copy templates to workspace root (don't overwrite existing): AGENTS.md, SOUL.md, USER.md, TOOLS.md, HEARTBEAT.md, IDENTITY.md
-
Create memory folders:
memory/people,memory/projects,memory/topics,memory/decisions -
Copy skills to
skills/ -
Optional: Add workflows — Ask if they want any workflows:
- "Do you want the email steward workflow?" → Manages inbox automatically
If yes:
- Copy
AGENT.mdfrom~/.openclaw-config/workflows/email-steward/toworkflows/email-steward/ - Create
workflows/email-steward/logs/directory - The workflow will interview them on first run to create
rules.md
-
Configure memory search — Required for semantic search to work. Ask: LM Studio (local, free, recommended) or OpenAI?
- LM Studio: Server on port 1234, model
lmstudio-community/embedding-gemma-300m-qat, configure gateway memorySearch.remote.baseUrl tohttp://127.0.0.1:1234/v1 - OpenAI: Get their API key, configure gateway memorySearch.remote.baseUrl to
https://api.openai.com/v1, modeltext-embedding-3-small - Verify it works: Create test file in memory/, index it, search for it, confirm it returns results, clean up
- LM Studio: Server on port 1234, model
-
Personalization — Ask and replace in templates:
{{USER_NAME}},{{ASSISTANT_NAME}},{{TIMEZONE}},{{PRIORITY_1}},{{PRIORITY_2}} -
Optional skills — These are optional. Ask about each one individually, only configure if they say yes:
- "Do you have a Limitless Pendant?" → If yes, get API key from app.limitless.ai → Settings → Developer
- "Do you use Fireflies.ai for meeting transcripts?" → If yes, get API key from app.fireflies.ai → Integrations → Fireflies API
- "Do you use Quo for business phone?" → If yes, get API key from my.quo.com → Settings → API
- "Do you want Parallel.ai for web search? (better results than built-in)" → If yes, get API key from platform.parallel.ai
Skip any they don't use. Don't assume they want all of them.
-
Health check setup — Ask: "Do you want automated health monitoring? It runs hourly via cron (Claude Code + Sonnet) to check your gateway and services, fix minor issues, and notify you of problems."
If yes:
- Ask: "Who should be notified if there's a problem?" →
mkdir -p ~/.openclawthen write their answer to~/.openclaw/health-check-admin(just the name, one line) - Resolve the claude CLI path with
which claude— use the full path in the cron job - Install the cron job (substitute CLAUDE_PATH with the resolved path):
0 0,7-23 * * * test -f "$HOME/.openclaw-config/devops/health-check.md" && flock -n "$HOME/.openclaw/health-check.lock" CLAUDE_PATH -p "Run health check" --model sonnet --append-system-prompt-file "$HOME/.openclaw-config/devops/health-check.md" --dangerously-skip-permissions --max-budget-usd 5.00 >> "$HOME/.openclaw/health-check.log" 2>&1 - Verify:
crontab -l | grep health-check - Do a test run: execute the claude command once and verify it produces output
- Tell them: runs hourly 7 AM–midnight, logs at
~/.openclaw/health-check.log
- Ask: "Who should be notified if there's a problem?" →
-
Track version in
.openclaw/installed-version -
Run diagnostics — Run
openclaw doctor --non-interactiveand review the output. This validates config integrity, credential health, security posture, skill eligibility, and memory search readiness. If doctor reports errors, fix them before proceeding. If it reports warnings (e.g., open DM policy), note them in the summary so the user can decide whether to address them. -
Summary — Tell them what's configured, include any doctor warnings
Update
Compare .openclaw/installed-version with ~/.openclaw-config/VERSION.
If newer: pull, update skills (safe to overwrite), update templates only if user hasn't modified them, update version file, report changes.
If user wants to force/overwrite: backup to .openclaw/backup/ first.
Workflows: Only update AGENT.md (the algorithm). Never touch rules.md, agent_notes.md, or logs/ — those belong to the user.
Status
Show installed version and skill versions. Fetch remote VERSION, report if updates available.