A collection of Agent Skills for AI coding assistants.

Agent Skills are an open format for giving agents new capabilities and expertise. Skills are portable across different agent products including Claude Code, Claude.ai, and other compatible agents.

Install any skill directly using npx:

# Install a skill
npx skills add linuxlewis/agent-skills/skills/agent-browser
npx skills add linuxlewis/agent-skills/skills/pr-responder
npx skills add linuxlewis/agent-skills/skills/ralph-runner
npx skills add linuxlewis/agent-skills/skills/openclaw-notify

# List installed skills
npx skills list

# Remove a skill
npx skills remove agent-browser

Alternatively, use the Claude Code plugin system:

# Add marketplace
claude plugin marketplace add linuxlewis/agent-skills

# Install skills
claude plugin install agent-browser@linuxlewis-agent-skills
claude plugin install pr-responder@linuxlewis-agent-skills
claude plugin install ralph-runner@linuxlewis-agent-skills
claude plugin install openclaw-notify@linuxlewis-agent-skills

Copy any skill folder from skills/ to your agent's skills directory.

Browser automation using the agent-browser CLI.

• Navigate websites and interact with page elements
• Take screenshots and PDFs
• Fill forms, click buttons, scrape content
• Session management for multi-step workflows

Requirements: npm install -g agent-browser

Review and respond to GitHub PR comments.

• Fetch and classify PR review comments
• Prioritize by actionability (bugs, security, style)
• Implement approved changes
• GitHub CLI commands reference

Requirements: GitHub CLI (gh) installed and authenticated

Command: /respond - Review PR comments for current branch

Run Ralph Wiggum autonomous coding loops.

• Execute PRD files with user stories
• Iterative implementation with fresh context per story
• Progress tracking and learnings
• Context file injection

Requirements: ralph-cli and Claude Code CLI

Notify OpenClaw gateway when background tasks complete.

• Send completion notifications back to dispatching session
• Report success, failure, or blocked status
• Simple one-command interface

Requirements: openclaw CLI installed, gateway running

Command: /notify <message> - Send notification to OpenClaw

agent-skills/
├── .claude-plugin/
│   └── marketplace.json        # Claude Code marketplace
├── skills/                      # Agent Skills compliant (portable)
│   ├── agent-browser/
│   │   ├── SKILL.md
│   │   └── references/
│   ├── pr-responder/
│   │   ├── SKILL.md
│   │   └── references/
│   ├── ralph-runner/
│   │   ├── SKILL.md
│   │   └── references/
│   └── openclaw-notify/
│       └── SKILL.md
├── plugins/                     # Claude Code specific (commands)
│   ├── pr-responder/
│   │   ├── .claude-plugin/plugin.json
│   │   └── commands/
│   └── openclaw-notify/
│       ├── .claude-plugin/plugin.json
│       └── commands/
└── README.md

Each skill follows the Agent Skills specification:

---
name: skill-name
description: What this skill does and when to use it.
license: MIT
compatibility: Required tools or environment
metadata:
  author: your-name
  version: "1.0.0"
---

# Skill Name

Instructions for the agent...

Skills use progressive disclosure for efficient context:

1. Metadata (~100 tokens) - Name and description loaded at startup
2. Instructions (< 500 lines) - Full SKILL.md loaded when activated
3. References (as needed) - Additional docs in references/ loaded on demand

1. Create a folder in skills/ with your skill name
2. Add SKILL.md with frontmatter and instructions
3. Optionally add references/, scripts/, or assets/ directories
4. Update marketplace.json to register with Claude Code

See agentskills.io for the complete specification.

• Agent Skills Specification
• Example Skills (Anthropic)
• agent-browser CLI
• Ralph Wiggum

MIT