OpenClaw Agent Skills

OpenClaw Agent Skills is an open-source repository published by the openclaw organization on GitHub under the MIT license. It provides a canonical set of reusable workflow skills for coding agents—such as Codex and Claude Code—so that common workflows like review closeout and remote validation can be written once and reused across every project without hand-copying long SKILL.md files.

What It Is

OpenClaw Agent Skills is a skill registry for AI coding agents. Each skill is a self-contained directory containing a SKILL.md file with YAML frontmatter and optional helper scripts. The repository ships an installer script (install-skills) that can symlink or copy skills into the agent's skill directory, making them immediately available to any agent that reads from that path. The project targets developers who run multiple agent-assisted repositories and want a single source of truth for shared operational workflows.

Included Skills

The repository currently ships five skills:

• agent-transcript — local-only, redacted PR/issue transcript provenance tracking
• autoreview — structured code-review closeout workflow with a helper test harness script
• crabbox — Crabbox/Testbox remote validation workflow for broad or CI-parity proof
• handoff — path-free prompt handoff workflow for delegating a task to another agent
• session-viewer — local searchable HTML viewer for agent session JSONL exports

How the Installer Works

The install-skills script supports several modes and targets. Developers can list available skills with --list, preview changes with --dry-run, install all skills or a named subset, specify a custom --target directory, and choose between symlink mode (default, best for local development) or copy mode (better for portable or locked-down setups). A --force flag replaces existing installed skills. For Codex, skills are symlinked into ~/.codex/skills; for Claude Code, into ~/.claude/skills.

Zero-Setup and Vendoring Model

The project distinguishes between live symlinks and vendored snapshots. Repos that need skills available without any setup can vendor a generated snapshot under .agents/skills/<name>. The README explicitly notes that vendored snapshots are distribution artifacts, not the source of truth: edits should happen in this repo first, then synced downstream. The autoreview skill is called out as a good candidate for zero-setup vendoring in flagship repos because review closeout is part of the standard contribution workflow.

Repository Layout and Validation

Each skill lives under skills/<name>/ and must contain a SKILL.md file; helper scripts belong inside that skill's own scripts/ subdirectory. A validate-skills script checks every skill for valid YAML frontmatter and required name and description fields. The validator also runs syntax checks on Ruby, Bash, Python, and Node scripts. The README notes that session exports can contain sensitive conversation data and that session-viewer HTML output should be treated as local/private unless separately redacted.

Current Status

The repository was created in May 2026 and last pushed in early June 2026, with 440 stars and 23 forks recorded on GitHub. It is actively maintained under the MIT license and is part of the broader openclaw ecosystem.