Skills

Overview

Skills are reusable instructions stored as SKILL.md files (YAML frontmatter plus markdown). Skills that allow model invocation are listed in the system prompt under <available_skills>, with the path of each file. The model reads that file with the read tool when the task matches the description.

A shared skills repository is coming soon, along with business use cases and more guides.

Discovery

Loop loads every SKILL.md it finds, in this order. The first skill with a given name wins.

~/.loop/agent/skills/
~/.agents/skills and ~/.claude/skills
When the project is trusted: .loop/skills in the working directory, then .agents/skills and .claude/skills in that directory and each ancestor
Extra directories listed in skills in settings.json

{
  "skills": ["~/shared-skills"]
}

~/.claude/skills and ~/.agents/skills are shared with other coding agents. A project .loop/skills directory is loaded only after /trust.

After you add or edit a skill, run /reload.

Using skills

Activate a skill

/skill:deploy
/skill:review src/api/

An activated skill stays in the catalog until /new, including skills that set disable-model-invocation. Text after the name is placed in the input buffer for your next message. For $1 and $@ substitution, use a prompt template. A disabled skill cannot be activated. Enable it first.

Turn skills on and off

/skills opens a picker. Space or Enter toggles the highlighted skill.

/skills list
/skills enable deploy
/skills disable pdf

on and off are aliases for enable and disable. ls is an alias for list. /skills list shows every discovered skill and whether it is enabled. A disabled skill is hidden from the model and from /skill: commands. If it was already active, it is deactivated. The choice is saved as disabledSkills in the global ~/.loop/agent/settings.json. A trusted project’s .loop/settings.json can add more names. Those names are combined with the global list.

{ "disabledSkills": ["pdf", "frontend-design"] }

Creating a skill

A skill is a directory with a SKILL.md file:

~/.loop/agent/skills/
├── deploy/
│   └── SKILL.md
└── review/
    └── SKILL.md

---
name: deploy
description: Deploy the application to production
---

## Instructions

You are a deployment specialist. Follow these steps:

1. Run the test suite: `cargo test`
2. Build the release binary: `cargo build --release`
3. Check for uncommitted changes: `git status`
4. If all checks pass, deploy using the deploy script

## Constraints

- Never deploy with failing tests
- Always create a git tag before deploying

Frontmatter

Field	Required	Description
`name`	No	Identifier for `/skill:name`. Defaults to the parent directory name
`description`	No	Shown in `/skills` and in the system-prompt catalog
`disable-model-invocation`	No	When `true`, the skill stays out of the catalog until `/skill:<name>`

---
name: review
description: Review code only when the user asks
disable-model-invocation: true
---

Review the paths the user names. Look for bugs, security issues, and style problems.

The body is the file the model reads. Write it as instructions for that task. Relative paths inside the file are resolved against the skill directory (the parent of SKILL.md).

Getting Started

Guides

Features

Overview

Discovery

Using skills

Activate a skill

Turn skills on and off

Creating a skill

Frontmatter

​Overview

​Discovery

​Using skills

​Activate a skill

​Turn skills on and off

​Creating a skill

​Frontmatter

Overview

Discovery

Using skills

Activate a skill

Turn skills on and off

Creating a skill

Frontmatter