Skip to main content
Skills are reusable instruction packs that inject domain-specific knowledge and behavior into agent sessions.

​
SKILL.md format

Each skill is a markdown file named SKILL.md with YAML frontmatter: 
---
name: react-best-practices
description: React component patterns and performance guidelines
---

When writing React components, follow these guidelines:

1. Prefer functional components with hooks over class components.
2. Use `useMemo` and `useCallback` for expensive computations and stable references.
3. Keep components small and focused on a single responsibility.

​
Frontmatter fields

FieldTypeRequiredDescription
namestringYesUnique skill name
descriptionstringYesDescription of what the skill provides
The markdown body becomes the skill’s instructions, injected into the agent’s context when the skill is active.

​
Discovery

Skills are discovered from multiple sources:

​
Local directories

0x0 scans for SKILL.md files in .zeroxzero/ directories:
  • Project: .zeroxzero/skill/**/SKILL.md or .zeroxzero/skills/**/SKILL.md
  • Global: ~/.zeroxzero/skill/**/SKILL.md or ~/.zeroxzero/skills/**/SKILL.md
Directories are walked upward from the current project directory to the worktree root.

​
Additional paths

Add extra skill directories via config: 
skills:
  paths:
    - ./vendor/skills
    - ~/my-skills
Relative paths are resolved from the project directory. Paths that do not exist are skipped with a warning.

​
Remote URLs

Fetch skills from remote servers: 
skills:
  urls:
    - https://example.com/.well-known/skills/
Remote skill discovery uses an index.json protocol: 
{
  "skills": [
    {
      "name": "react-best-practices",
      "description": "React component patterns and performance guidelines",
      "files": ["SKILL.md", "examples.md"]
    }
  ]
}
Files are fetched from {base_url}/{skill_name}/{file} and cached locally in ~/.cache/zeroxzero/skills/. Only skills containing a SKILL.md file are loaded.

​
Duplicate handling

If two skills share the same name, the later-discovered skill overwrites the earlier one. A warning is logged when this happens.

​
Permission integration

Skill directories are automatically added to the permission system as allowed paths for the external_directory permission. This means agents can read skill files without additional permission configuration.

​
Debugging

List all discovered skills: 
0x0 debug skill

​
See also