skill-generator

Skill Generator

Use this skill when helping a user create, revise, review, validate, or package an OpenClaw skill.

Operating Rules

• •Treat OpenClaw official documentation as authoritative.
• •Keep SKILL.md concise, operational, and easy to review.
• •Separate skill instructions from external capabilities.
• •Do not imply a skill grants access to tools, binaries, APIs, credentials, files, or permissions.
• •Identify required tools, binaries, APIs, credentials, environment variables, config keys, network access, and filesystem access separately.
• •Never embed secrets, tokens, API keys, private keys, session cookies, personal data, or SecretRef values in skill files.
• •Treat third-party skills as untrusted source material until reviewed.
• •Prefer simple folder structures and plain Markdown over complex automation.
• •Recommend scripts only when they improve repeatability, validation, or safety.
• •Preserve Skill Workshop as the governed proposal mechanism when agent-generated skill changes should be reviewed before installation.

Workflow

When creating or improving a skill:

1. Clarify the skill purpose, user, trigger, and non-goals.
2. Identify required external capabilities and permission boundaries.
3. Choose the smallest reviewable folder structure.
4. Draft or revise SKILL.md.
5. Add references/ only when supporting material would make SKILL.md too long.
6. Add scripts/ only when deterministic automation is safer than prose.
7. Add examples/ only when examples improve review or testing.
8. Run a security and maintainability review.
9. Prepare the skill for human review before installation or publication.

Reference Files

Use these reference files when needed:

• •{baseDir}/references/skill-design-workflow.md
• •{baseDir}/references/review-checklist.md
• •{baseDir}/references/security-guidelines.md
• •{baseDir}/references/third-party-review.md
• •{baseDir}/references/skill-workshop-flow.md

Output Style

For small requests, provide concise guidance.

For implementation work, provide copy-ready files or commands, grouped by filename.

When reviewing a skill, classify findings as:

• •Verified
• •Stated by user
• •Inference
• •Open question
• •Recommendation

Do not present draft, reviewed, approved, installed, deprecated, or unsafe skills as the same maturity level.