Skills

Skills are named instruction sets the agent can load when a task calls for them, such as a testing approach or a release checklist. They persist across chats and are stored per project.

How it works#

1. At chat start, every skill is listed in the system prompt as one line: role/name: summary.
2. When a skill is relevant, the agent calls GetSkill(name) and receives the full content as a tool result.
3. The agent follows it for the task at hand.

Only summaries cost tokens up front. Full content is loaded on demand.

You can also force a skill into the conversation yourself with /use <name>.

Characteristics#

• Per role, by name. A role can have many skills, e.g. code/testing, code/git-workflow. GetSkill and SetSkill always act on the agent's own role.
• Stored in the project. Skills live in .raggie/.raggie.chat, so each project has its own set.
• Agent-authored, with consent. The agent can create or update a skill with SetSkill. You are shown the full content and asked before anything is saved.
• Summary source. The description from the frontmatter if there is one, otherwise the first line of the content (up to 200 characters).

Agent Skills format#

Raggie reads and writes the Agent Skills format: a folder per skill with a SKILL.md file.

my-skills/
├── web-testing/
│   └── SKILL.md
└── git-workflow/
    └── SKILL.md

---
name: web-testing
description: Systematic approach to testing web applications
---

# Web Testing

When asked to test a web application:
1. Start with unit tests for individual components
2. Write integration tests for API endpoints

name becomes the skill name (the folder name is the fallback) and description becomes the summary. Other frontmatter fields are kept as is.

Only SKILL.md is imported. The optional scripts/, references/ and assets/ folders from the spec are not loaded.

Auto-discovery#

At chat start Raggie scans these folders in the project root and imports any skill it does not already have:

1. .skills/
2. skills/
3. .claude/skills/

Existing skills with the same name are never overwritten. Drop a skill folder into one of these directories and it is available on the next chat, or immediately with /reload-skills.

.agents/skills/ is not scanned automatically. Import from it with /importSkills.

In-chat commands#

CLI#

Run these from the project directory.

raggie skill code                                   # interactive menu
raggie skill code --show --name testing             # print one skill
raggie skill code --import-skill my-skill.md --name testing
raggie skill code --export-skill backup.md --name testing
raggie skill code --delete --name testing
raggie skill --list-all

# Agent Skills folders
raggie skill code --import-skill-dir ./my-skills/              # a folder of skill folders
raggie skill code --import-skill-dir ./my-skills/web-testing/  # one skill folder
raggie skill code --export-skill-dir ./exported/ --name web-testing

When importing a Markdown file that has a name in its frontmatter, that name wins over --name. Exporting to a folder generates frontmatter if the skill has none.

Where skills sit in the prompt#

1. Role system prompt
2. Date, working directory, host info
3. Skill summaries
4. AGENTS.md

AGENTS.md comes last, so use it for rules that must always apply, and skills for knowledge that is only needed sometimes.