September 26, 2026

OpenClaw Agent Skills
Reusable, shared workflow skills for coding agents like Codex and Claude Code.
git clone https://github.com/openclaw/agent-skills.git
cd agent-skills
scripts/install-skillsProduct facts
- Type
- AI Skill & Agent
- Category
- AI & Machine Learning › Prompt Engineering & Skill Management
- Pricing
- Open source
- Platforms
- Claude Code
- License
- MIT
- Launched
- Sep 26, 2026
- Listed
- Sep 26, 2026
- Website
- github.com
Overview
## Introduction OpenClaw Agent Skills is a public specification repository hosting shared skills for coding agents. Its core goal is to let general-purpose workflows be written once and reused everywhere, avoiding the need to manually copy lengthy `SKILL.md` files into every repository. ## Included Skills The repository currently includes the following skills: - **agent-transcript**: Local-only, redacted PR/Issue transcript provenance. - **autoreview**: Structured code review and wrap-up workflows with helper scripts. - **behavior-validator**: Validates user-visible behavior against contracts without inspecting source code. - **beam**: Publishes local coding sessions to a read-only OpenClaw catalog in a self-contained, authenticated, and redacted manner. - **crabbox**: Crabbox/Testbox remote validation workflows for broad verification or CI-equivalent proof. - **handoff**: Path-independent prompt handoff workflow for delegating tasks to another agent. - **readme-standard**: README structure, badge row, tone, and validation gate specifications for the steipete/openclaw repository. - **session-viewer**: A local, searchable HTML viewer for agent session JSONL files. ## Installation & Usage Skills are installed via the `scripts/install-skills` script provided in the repository. It supports listing available skills, dry-run previews, full installation, selective installation, and installing to a specified target directory. Symlinking is used by default so changes take effect immediately during local development; you can also use `--mode copy` for portable or restricted environments. For Codex, symlink the repository to `~/.codex/skills`; for Claude Code, link it to `~/.claude/skills`. ## Use Cases Ideal for teams or individuals using coding agents within the OpenClaw ecosystem to standardize common processes such as code review, behavior validation, and task handoffs. Product-specific skills should remain in their respective repositories, while general-purpose helper skills are managed centrally here.
Key features
- autoreview: Provides structured code review and wrap-up workflows along with helper scripts.
- behavior-validator: Validates user-visible behavior against contracts without inspecting source code.
- crabbox: Remote validation workflows for broad verification or CI-equivalent proof.
- handoff: A path-independent prompt handoff workflow for delegating tasks to other agents.
- agent-transcript: Local-only, redacted provenance for PR or Issue transcripts.
- beam: Publishes local coding sessions to a read-only catalog in a self-contained, authenticated, and redacted way.
- readme-standard: Standardizes README structure, badge rows, tone, and validation gates.
- session-viewer: A local, searchable HTML viewer for agent session JSONL files.
- Flexible install script: Supports dry-run, custom directories, symlinks, or copy modes.
Releases
Notify meQ&A
- How do I install skills into Codex?AI-generated
- After cloning the repository, run `mkdir -p ~/.codex/skills` and then `ln -sfn "$(pwd)/skills" ~/.codex/skills/agent-skills` to symlink the skills directory into Codex's skills path.
- How do I install only specific skills?AI-generated
- Simply specify the skill names after the install command, for example `scripts/install-skills autoreview crabbox`.
- What is the difference between symlink and copy mode?AI-generated
- Symlinks are ideal for local development, as changes to the repository take effect immediately; copy mode suits portable or restricted environments. Node-based skills declare their module format locally, so copied helper scripts also work in CommonJS projects without installing dependencies.
- What environment is required for development checks?AI-generated
- Development checks use Python 3.14 and Node.js 26 to stay consistent with CI. You need to create a virtual environment, install the dependencies from requirements-dev.txt, and run npm ci.
- Are session-viewer exports safe?AI-generated
- Session exports may contain sensitive conversation data. Unless separately redacted and reviewed, treat HTML generated by session-viewer as local or private output.
This product has no owner yet
It was listed by the platform or its author left. If it is your work, prove your relationship to become the owner.
Claim this productAlternatives

YouTube Skills for AI Agents
Equips AI agents with YouTube transcript retrieval, video search, channel browsing, and playlist extraction.

libtv-skills
面向 AI Agent 的技能包,通过 LibLib.tv OpenAPI 提供生图、生视频等 AIGC 能力。
WebGPU Three.js TSL Skill
Provides knowledge and rules for developing WebGPU Three.js apps with TSL in Claude Code and Cursor.

screenwriting-skills
为编剧和戏剧创作者提供 26 个 AI Agent 技能,覆盖电影、剧集与舞台剧本写作。

Vibe Security
一个 Agent Skill,用于审计 AI 辅助编写的应用中常见的安全漏洞模式。

Higgsfield AI Skills
为 AI 编程代理提供图像、视频生成及品牌视觉系统构建能力的技能集。

guard-skills
Quality-check gates for AI coding agents covering code, tests, and docs—catch common generation flaws before commit.

Kotlin Agent Skills
A collection of AI agent skills following the Agent Skills standard for Kotlin projects.

AWS Agent Skills
为 Claude Code 和 Codex 提供覆盖 18 项 AWS 服务的本地化工程知识技能包

tutor-skills
A Claude Code skill that turns PDFs, documents, and codebases into an Obsidian study vault and tracks concept mastery through interactive quizzes.
The product content on this page was translated by AI from the author's original.