openclaw

Watch Skill in OpenClaw

Status: doc-verified ☑ — matches the openclaw.ai skills docs; not executed here.

OpenClaw's native extension surface is the same one we ship: directories of SKILL.md files. No MCP config needed — the CLI-wrapping skills work as-is.

Install

# puts `watch-skill` on PATH, which the skills call
uv tool install "watch-skill[standard]"   # or: pipx install "watch-skill[standard]"
watch-skill doctor

Configure

Point OpenClaw's skill discovery at the skills library (it scans any configured root for SKILL.md, up to 6 levels deep). In your OpenClaw config, under skills.load:

{
  "skills": {
    "load": {
      "extraDirs": ["C:\\path\\to\\watch-skill\\adapters\\claude-skill\\skills"]
    }
  }
}

That picks up all of them: watching-videos, asking-with-evidence, the-loop, learning-from-mistakes, extracting-structure, video-memory, sharing-results, configuring-vision, recovering-from-errors, plus /watch.

Smoke test (3 steps)

1. Ask: "Watch https://www.youtube.com/watch?v=aqz-KE-bpKQ — what happens at 0:10?" — the watching-videos skill should fire and run watch-skill watch.
2. Check the reply cites t=MM:SS timestamps from the report.
3. Follow up: "what color is the bird?" — the asking-with-evidence skill should run watch-skill ask, not re-watch.

Notes

• Our frontmatter carries name, description, version, license, allowed-tools. OpenClaw ignores keys it doesn't know; the trigger description is what matters.
• The skills call the watch-skill CLI, so the uv tool install . step (or pip install watch-skill) is what makes them portable here.