openclaw/agent-skills

OpenClaw Agent Skills – What It Is

OpenClaw Agent Skills is a small, focused library that bundles reusable “skills” (i.e. ready‑made workflows and helper scripts) for AI‑powered coding agents that work on OpenClaw projects. Rather than copy long SKILL.md files into every repository, teams can install a single shared skill set and have their agents invoke the same, centrally‑maintained logic.

Core Idea

• Shared, version‑controlled workflows – each skill lives in its own folder under skills/ and contains a SKILL.md descriptor plus any scripts needed to run the workflow.
• One‑time install – a tiny Python installer (scripts/install-skills) can symlink or copy the skills into a user’s ~/.agents/skills (or Codex/Claude) directory, making them instantly available to any repository.
• Agent‑agnostic – the repo mentions Codex and Claude, but the skills are just files and scripts; any LLM‑driven coding agent that follows the OpenClaw “skill” convention can use them.

What’s Inside?

Skill	Purpose
agent‑transcript	Keeps a redacted, local‑only transcript of PR/issue discussions for provenance tracking.
autoreview	A structured code‑review/close‑out workflow with a helper script that agents can run automatically.
behavior‑validator	Validates the observable behavior of a change against a contract without looking at the source code.
beam	Publishes a coding session (redacted) to a read‑only OpenClaw catalog, with authentication.
crabbox	Remote validation workflow used for broad or CI‑parity proof of work.
handoff	Passes a task from one agent to another via a prompt‑based handoff.
readme‑standard	Enforces a standard README layout, badge row, tone, and verification gates for OpenClaw repos.
session‑viewer	Generates a searchable local HTML view of an agent’s JSONL session export.
team‑handoff	Starts a seeded work‑tree session on a shared OpenClaw Gateway as the current user.

How to Use It

1. Clone the repo.
2. Run scripts/install-skills (with options like --list, --dry-run, --target, --mode copy, --force).
3. The installer creates symlinks (or copies) of the selected skill folders into your agent‑skill directory (e.g., ~/.codex/skills or ~/.claude/skills).
4. Repositories reference the installed skill, e.g., by having a .agents/skills/autoreview/SKILL.md entry point that points to the shared version.

Development & Validation

• Uses Python 3.14 and Node 26 for CI checks.
• scripts/check-skills runs front‑matter validation, syntax checks, Python/Node tests, and TypeScript type‑checking for the session viewer.
• A lighter scripts/validate-skills only checks the YAML front‑matter of each SKILL.md.

Release Process

Changes are made in this canonical repo, then downstream repositories can vendor a snapshot of the needed skills (under .agents/skills/<name>). Snapshots are distribution artifacts only; the source of truth stays here.

License

MIT – free to use, modify, and redistribute.

---

Bottom line: OpenClaw Agent Skills is a practical toolkit that lets AI coding agents share common, maintainable workflows across many repositories, reducing duplication and keeping review/validation processes consistent.