Skills

Skills are on-demand capability packs: workflows, scripts, and reference docs. AutoC follows the Agent Skills standard.

Quickstart

Browse Skill plaza, then in AutoC:

/skill-install <skill-id>

Or in the shell:

bash

autoc skill install <skill-id>

Installs to global (~/.autoc/agent/skills/) or project (.autoc/skills/). /reload, then /skill:name.

Download ZIP → extract to a skills directory above → /reload.

Scope	Path
Global	`~/.autoc/agent/skills/`, `~/.agents/skills/`
Project	`.autoc/skills/`, `.agents/skills/` (up to git root)
Settings	`skills` array in settings.json
CLI	`--skill <path>`

--no-skills disables auto-discovery; explicit --skill paths still load.

Reuse Claude Code / Codex skill dirs:

json

{
  "skills": ["~/.claude/skills", "~/.codex/skills"]
}

Form	Purpose
`/skill:name`	Load and run skill
`/skill:name args`	Args appended as `User: <args>`

Enable in /settings or enableSkillCommands: true in settings.json.

my-skill/
├── SKILL.md              # required
├── scripts/
└── references/

SKILL.md needs frontmatter (name, description required) and body. Paths relative to the skill directory.

markdown

---
name: my-skill
description: What it does and when to use it. Be specific.
---

# My skill

## Usage
...

Field	Notes
`name`	lowercase, hyphens, matches folder, max 64 chars
`description`	max 1024 chars, drives when to load
`disable-model-invocation`	`true` → user must invoke via `/skill:name`

Spec: Agent Skills specification. Most validation issues are warnings; skills without description are not loaded.

Security

Skills can instruct the model to run arbitrary actions. Review before use.