OpenClaw Agent Skills - Setup Guide¶
Source: openclaw/agent-skills GitHub: openclaw/agent-skills Skills: 8 skills · ~2.0K total installs Category: Agent Operations First Seen: August 19, 2026 sweep Quality Tier: 🟡 Beta - official OpenClaw org repository (1,057 GitHub stars, actively maintained - last commit August 18, 2026); install counts still growing
This is the public canonical source for shared agent workflows from the OpenClaw organization itself. The goal stated in the repo: write a workflow once, reuse it everywhere, and avoid hand-copying long SKILL.md files across repos. The 8 skills cluster into three families - review (autoreview, behavior-validator, crabbox), handoff/context (handoff, agent-transcript), and sessions/documentation (session-viewer, beam, readme-standard). autoreview is the flagship at 1,388 installs, a structured multi-engine code review workflow (Codex, Claude, Amp, Pi, Kimi) that reports blocking issues only when explicitly requested.
Installation¶
Install everything via the bundled installer (lists, dry-runs, and selective installs supported):
git clone https://github.com/openclaw/agent-skills.git
cd agent-skills
scripts/install-skills --list # see available skills
scripts/install-skills --dry-run # preview without changes
scripts/install-skills # all skills, default agent skill dir
scripts/install-skills autoreview crabbox # selective
scripts/install-skills --target ~/.codex/skills autoreview # custom target
scripts/install-skills --mode copy --target ~/.agents/skills # copies, not symlinks
Or via the skills.sh marketplace (skills are indexed there):
npx skills add openclaw/agent-skills
Prerequisites¶
| Requirement | Details |
|---|---|
| Node.js + npx | For the skills.sh installer path |
| An agent runtime | Skills assume a Codex/Claude/OpenClaw/Pi-style SKILL.md runtime |
| Codex (optional) | autoreview's default review engine (gpt-5.6-sol, high reasoning) |
| GitHub CLI or git | agent-transcript writes PR/issue bodies; crabbox runs remote validation |
What It Provides¶
| Skill Group | Skills | Purpose |
|---|---|---|
| Review | autoreview, behavior-validator, crabbox | Structured code review closeout, source-blind behavior validation against a contract, and Crabbox/Testbox remote proof (Linux/macOS/Windows/WSL2) |
| Handoff | handoff, agent-transcript | Clipboard-ready prompt handoff to another agent; redacted local-only PR/issue transcript provenance |
| Sessions | session-viewer, beam | Local searchable HTML viewer for agent session JSONL; authenticated redacted publication of sessions to an OpenClaw catalog |
| Documentation | readme-standard | House README structure, badge row, tone, and verification gates for steipete/openclaw repos |
Top skills by installs: autoreview (1,388), agent-transcript (355), session-viewer (66), handoff (62), behavior-validator (49), crabbox (42).
Quick Start¶
- Clone and list:
git clone https://github.com/openclaw/agent-skills.git && cd agent-skills && scripts/install-skills --list - Install the flagship:
scripts/install-skills autoreview - For review closeout, say "run autoreview on this change" - it reports P0 blocking issues by default
- For handoffs, say "write a handoff for the next agent" - it builds a standalone prompt and copies it to the clipboard
CorpusIQ Use Cases¶
| Use Case | How |
|---|---|
| Agent code review closeout | autoreview's P0-only default output maps to our verify-before-assertion discipline - blocking issues only, advisory output, no blind apply |
| Multi-agent handoffs | handoff builds path-free standalone prompts for fresh agents - matches our multi-agent orchestration and delegation patterns |
| Session forensics | session-viewer renders session JSONL as searchable single-file HTML - useful for auditing agent runs and sharing context |
| PR/issue transparency | agent-transcript attaches sanitized agent transcripts to PRs (redacted, fail-closed on secrets) - matches our public-content sanitization rules |
| Behavior validation | behavior-validator checks user-visible behavior against a contract without source inspection - useful for testing skills against their documented behavior |
Integration with CorpusIQ¶
CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.
This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.
Limitations / Verification¶
- Below the 20K install guide bar - drafted on official-org authority (the OpenClaw org itself, 1,057 GitHub stars, last commit August 18, 2026)
- Skills are Codex/OpenClaw-first; Hermes usage works via SKILL.md conventions but the installer targets generic agent skill directories
- agent-transcript is explicitly best-effort and local-only; PR/issue creation continues without it
- beam requires an authenticated OpenClaw receiver endpoint
npx skills add openclaw/agent-skills # verify install works
Related¶
Powered by CorpusIQ