ZeroClaw Skills Index

A skill registry for ZeroClaw — AI agent skills, tools, and workflows. Skills follow the agentskills.io open specification.

What we audit, what we don't

This registry is not a vetted security boundary. CI runs structure validation and pattern-matching scans (secret detection, dangerous-shape regexes, file-policy checks) on every PR. We do not audit skill code for security or correctness, and pattern matching does not catch novel or obfuscated misuse. Runtime behavior — network calls, package installs, files written, commands run — is the user's responsibility to evaluate before installing.

Run zeroclaw skills audit <name> and read the skill's SKILL.md before installing anything you don't recognize.

Trust tiers

The tags field in registry.json carries the trust signal:

• Official — authored and maintained by zeroclaw-labs. Same contribution rules as everything else, but the author is us.
• Community — submissions from external contributors. The registry lists them and CI checks structure; we make no representation about the author's identity, intent, or the skill's runtime behavior. Treat them as you would any third-party code.

Install a skill

zeroclaw skills install <skill-name>

Browse skills

Visit the Skills Hub to search and discover skills.

---

Contributing a skill

Anyone can publish a skill. Fork this repo, add your skill, and open a pull request. Two automated CI checks (structure validation + pattern-matching scans) must pass before a maintainer reviews. Passing CI is a baseline filter, not an endorsement of the skill's runtime behavior — see What we audit, what we don't.

Step 1 — Fork & clone

gh repo fork zeroclaw-labs/zeroclaw-skills --clone
cd zeroclaw-skills

Step 2 — Create your skill folder

skills/my-skill/
├── SKILL.md        # Required — YAML frontmatter + agent instructions
└── README.md       # Required — documentation for users

Your folder name must match the name field in the SKILL.md frontmatter.

About author: community: A handful of legacy entries in registry.json carry "community" as the author. These are placeholder values for skills whose original contributor couldn't be resolved from git history, and they are being phased out as new community-authored skills replace them. New contributions must use a real GitHub login — the placeholder is not accepted for new submissions.

Categories

Accepted licenses (SPDX identifiers)

MIT · Apache-2.0 · CC-BY-4.0 · CC-BY-SA-4.0 · CC0-1.0 · BSD-2-Clause · BSD-3-Clause · ISC · Unlicense

Permissions

Only request what your skill actually needs. Dangerous combinations are blocked by CI.

Blocked combos: shell_exec + any network permission, file_write + shell_exec. These combinations allow arbitrary code execution with data exfiltration and will be rejected automatically.

Step 3 — Write SKILL.md

SKILL.md is the single source of metadata. It starts with YAML frontmatter followed by the agent instructions, per the agentskills.io specification. All fields live at the top level of the YAML — no metadata: sub-block.

Frontmatter

---
name: my-skill
description: >-
  What your skill does and when to use it. Include keywords that help
  agents identify relevant tasks.
version: "0.1.0"
author: your-github-username
license: MIT
category: tools
tags:
  - Community
permissions:
  - file_read
---

Instructions body

Below the frontmatter, write the prompt/instructions that ZeroClaw feeds to the AI agent when your skill is activated.

---
name: my-skill
description: >-
  Brief description of what the skill does and when to use it.
version: "0.1.0"
author: your-github-username
license: MIT
category: tools
tags:
  - Community
permissions:
  - file_read
---

# My Skill

You are a [role]. When given a task:

1. First step
2. Second step
3. Third step

## Rules
- Be specific about constraints
- Define the output format

Step 4 — Write README.md

Documentation for users browsing the registry. Include:

• What the skill does
• Install command (zeroclaw skills install my-skill)
• What permissions it needs and why
• Example usage

Step 5 — Add to registry.json

Add an entry to the skills array in registry.json:

{
  "name": "my-skill",
  "version": "0.1.0",
  "author": "your-github-username",
  "description": "What your skill does in one sentence.",
  "category": "tools",
  "tags": ["Community"],
  "downloads": 0,
  "stars": 0
}

Step 6 — Open a pull request

git checkout -b add-my-skill
git add skills/my-skill/ registry.json
git commit -m "feat: add my-skill"
git push origin add-my-skill
gh pr create --title "Add my-skill" --body "Adds the my-skill skill for [brief purpose]."

---

What happens when you open a PR

Two CI workflows run automatically:

Validation (validate.yml)

• registry.json is valid JSON
• Every skill folder has a non-empty SKILL.md with valid YAML frontmatter
• All required top-level frontmatter fields are present (name, description, version, author, license, category, tags, permissions)
• license is a valid SPDX identifier
• Folder names match the frontmatter name field
• registry.json and skills/ folders are in sync

Security scan (security-scan.yml)

• Secret detection — scans for leaked API keys, tokens, and passwords
• File policy — only .md, .json, .wasm, .yml, .yaml files allowed; max 500KB; no symlinks or hidden files
• Content scanning — checks for prompt injection patterns, dangerous shell commands, suspicious URLs, credential leaks, and obfuscated payloads
• WASM analysis — blocks dangerous system imports in .wasm binaries
• Registry integrity — validates semver, categories, no path traversal, no external URLs
• Permission audit — flags dangerous permission combinations (see Permissions above)
• URL safety — checks URLs against Google Safe Browsing (when configured)

Both checks must pass. If a check fails, a comment will explain what went wrong.

---

Reporting a malicious or compromised skill

This is a removal and advisory process, not a vetting process. Skills are scanned at PR time but not continuously audited. If a published skill is acting maliciously — for example, exfiltrating data, executing unexpected shell commands, or fetching and running remote code at runtime — please report it.

How to report

Open a GitHub Issue in this repo with a title prefixed [security] followed by the skill name, e.g. [security] my-skill. Include the skill version, what you observed, and any logs or reproduction steps you can share.

What we'll do

• Triage within 48 hours of the report being filed.
• On confirmation, remove the skill from registry.json immediately.
• Move the skill directory to archive/<name>-<sha>/, where <sha> is the commit being archived.
• File a security advisory via this repo's Security tab.

What we won't do

• Conduct forensic analysis of compromised contributor accounts.
• Contact upstream services (package registries, hosting providers, etc.) on a user's behalf.
• Guarantee a response time outside the stated 48-hour triage window.

What users can do

• Pin skill versions when installing: zeroclaw skills install <name>@<version>.
• Run zeroclaw skills audit before installing a new skill.
• Watch this repo's Security tab for advisories.

---

License

MIT