Skills

A skill is a folder with a SKILL.md file and optional supporting files. Open Interpreter reads skill metadata first, then loads the full skill only when the request matches.

Folder Shape

cut-release/
├── SKILL.md
├── scripts/
├── references/
└── assets/

SKILL.md is required. The other directories are optional.

Minimal Skill

---
name: cut-release
description: Prepare a release by testing, updating changelog, and tagging.
---

When asked to cut a release:

1. Run the test suite.
2. Update the changelog.
3. Bump the version according to semver.
4. Prepare the commit and tag, but ask before publishing.

The description controls when the skill is selected, so make it specific.

Locations

Local skills take priority over personal and bundled skills when names collide.

Use the .agents/skills locations for new skills. They are shared, tool-neutral directories, so another compatible agent can use the same skill without an Open Interpreter import or copy.

Open Interpreter also reads ~/.openinterpreter/skills/ as a legacy compatibility fallback. Do not use that product-specific directory for new user-authored skills. Bundled internal skills may still be cached under the Open Interpreter home because they are product assets rather than portable user data.

Bundled Skill Updates

Open Interpreter owns the bundled skill cache at $INTERPRETER_HOME/skills/.system/ (normally ~/.openinterpreter/skills/.system/). The runtime fingerprints the complete embedded bundle and refreshes that directory when a newly installed Open Interpreter version ships changed or retired skills. This happens when the updated runtime starts a session, including interactive, exec, and app-server sessions.

Treat .system/ as read-only: an Open Interpreter update can replace changes inside it. User skills, host-application skills, and other sibling directories under skills/ are outside Open Interpreter's managed namespace and are not changed by this refresh.

What to Put in a Skill

Use skills for repeatable procedures:

• Release checklists
• Internal report generation
• Repo-specific migration workflows
• Design or review standards
• Commands that need fixed ordering

Keep SKILL.md concise. Put long references in references/, runnable helpers in scripts/, and templates in assets/.

Tool and Approval Behavior

Skill scripts run through normal sandbox and approval controls. A skill should describe what the script does and when it is appropriate to run, but it should not rely on bypassing permissions.

Browse Skills

Inside the TUI:

/skills