| name | claw-config-skill-creator |
| description | Create or review OpenClaw skills for the claw-config agent. Default mode drafts a new skill (proposal-only) with docs-first diagnosis, change control, and claw-config safety constraints; review mode audits an existing skill for correctness and safety. |
| compatibility | OpenClaw |
| metadata | {"author":"claw-config","version":"1.1","openclaw":{"emoji":"🧱"}} |
claw-config-skill-creator
A blueprint skill for creating claw-config-specific skills: conservative config operations with docs-first diagnosis, change control, and consistency guarantees.
Interface
Default: create (proposal-only)
/claw-config-skill-creator <new-skill-name>
Produces a complete PROPOSAL (file tree + full contents) for a new skill.
Review: audit an existing skill
/claw-config-skill-creator review <skill-path-or-name>
Outputs a review report: issues, risks, and concrete edits.
Hard parsing rules
- First token after
/claw-config-skill-creator is either:
review, or
<new-skill-name>
- Any other token is unsupported → show help (in
references/HELP.md) and ask the user to restate.
Output contract
Create mode
Must output:
- Target skill name + intended trigger description
- Proposed file tree
- Full contents of every file
- Any open questions (only if blocking)
Default is proposal-only (no writes).
Review mode
Must output:
- Summary: pass/fail + risk level
- Findings grouped by category (docs-first, routing vs activation, apply gating, public hygiene)
- Suggested patch snippets
Mental model (must be embedded in created skills)
A claw-config skill must clearly separate:
- Routing:
bindings chooses which agent receives a message.
- Activation / access:
channels.* gates whether replies happen (mentions/allowlists/topic overrides, etc.).
Docs-first (no fixed doc lists)
Created skills should not embed a curated list of docs pages.
Instead, they must include a short Docs-first protocol:
- Start from official docs.
- Find the right page by searching for the exact config keys / error strings involved.
- If docs are ambiguous or version-sensitive, confirm in source and validate with a minimal repro.
A good skill teaches how to discover the right doc, not which doc to memorize.
Guardrails (non-negotiable)
- Docs-first for deep issues. No guessing.
- Claw-config boundaries: stay in config + ops; keep secrets redacted by default.
- Plan-first defaults: for any skill that can change config, default behavior must be plan/proposal; apply must be a gated follow-up step.
- Rollback + verification + report: require rollback pointer and a post-change report (what/verify/effective/risk).
- Public hygiene: no private absolute paths; no version-sensitive routing logic copied into skills—link to official docs instead.
References (must-read)
references/ROLE_AND_BOUNDARIES.md
references/CHANGE_WORKFLOW.md
references/TOOLING_AND_SECURITY_MODEL.md
references/ARCHETYPES.md
references/CREATE_TEMPLATE.md
references/REVIEW_CHECKLIST.md
references/PUBLIC_HYGIENE.md
references/HELP.md