Skills for Agents

Skills repository for OpenClaw agents. Contains foundational and specialized skills that agents can import and use.

Structure

├── Foundational/     — Base skills that document APIs, services, and tools. Other skills reference these.
├── Specialized/      — Niche skills that build on foundational skills for specific workflows.
├── COORDINATION.md   — Communication channel between builder and reviewer agents.
└── Plan.MD           — Progress tracker for skill development.

How This Repo Works

Two agents collaborate through this repo:

Builder Agent (Douge)

• Researches API docs thoroughly, builds foundational and specialized skills
• Pushes skills to feature branches (skill/<name>)
• Creates PRs with test instructions
• Labels PRs needs-testing when ready for review
• Reads reviewer feedback and fixes issues on failed tests

Reviewer/Tester Agent (Choug)

• Watches for PRs labeled needs-testing on cron (every 30 min)
• Pulls the branch and imports the skill into the workspace skills folder
• Spawns a clean sub-agent session (no prior context of the skill) to test it naturally
• Analyzes the sub-agent's session results for correctness
• Comments on the PR with results:
  • Pass: What was tested, what worked, evidence (command outputs, API responses)
  • Fail: What broke, exact error, what was confusing, suggested fixes
• Labels PR tested-pass or tested-fail
• Checks COORDINATION.md for additional context or instructions from the builder
• Has final accept/reject authority on skill quality before Kyle merges

Human (Kyle)

• Drops skill requests to the builder agent
• Reviews and merges PRs that pass testing
• Final authority on all merges

PR Workflow

Builder creates branch → pushes skill → opens PR (needs-testing)
    ↓
Reviewer pulls branch → tests in clean session → comments results
    ↓
Pass → label: tested-pass → Kyle reviews & merges
Fail → label: tested-fail → Builder reads feedback → fixes → re-pushes → cycle repeats

Reviewing a Skill — Guidelines for the Tester Agent

When you pick up a PR labeled needs-testing:

1. Read the PR description for test instructions and what the skill does
2. Check COORDINATION.md for any additional context from the builder
3. Import the skill into your environment
4. Open a new, clean session — do NOT test in the same session where you imported it
5. Try to trigger the skill naturally — use the kinds of prompts a real user would. The skill's description tells you what should trigger it.
6. For API skills: Actually call the endpoints if you have the API key. Verify the request format, headers, and response parsing are correct.
7. For setup skills: Run the setup script. Verify it installs what it claims.
8. Report on the PR:
   • What you tested (specific prompts/commands)
   • What worked
   • What failed (with exact errors)
   • Whether the skill instructions were clear or confusing
   • Any missing information you had to guess at
9. Also update COORDINATION.md with your test results summary

Skill Quality Standards

Every skill in this repo must:

• Have proper YAML frontmatter (name, description with trigger scenarios, metadata)
• Folder name matches the name field
• Description includes specific numbered trigger scenarios
• Body is under 500 lines
• Examples use real, working endpoints
• References are linked from the body with guidance on when to read them
• No extraneous files (README, CHANGELOG inside skill folders)

Labels

Label	Meaning
needs-testing	Builder finished, ready for reviewer
tested-pass	Reviewer verified it works
tested-fail	Reviewer found issues (see PR comments)
foundational	Base API/service skill
specialized	Builds on foundational skills