> ## Documentation Index
> Fetch the complete documentation index at: https://docs.selftune.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Agent skill

> Install SelfTune as an agent skill and operate it with intent-level requests.

The agent skill is the recommended interactive interface. It teaches your
agent how to inspect SelfTune state, choose the appropriate workflow, invoke the
local runtime, verify results, and stop at consent boundaries.

## Install

```bash theme={null}
npx skills add selftune-dev/selftune
```

Then start a fresh agent session if necessary and say:

```text theme={null}
Initialize SelfTune. Inspect this environment first, keep setup local, and ask
before installing hooks or importing historical sessions.
```

## Use intent-level requests

* “Show me which skills need attention.”
* “Explain why this skill has no data.”
* “Preview an improvement for the research skill.”
* “Create and verify a skill for release-note reviews.”
* “Prepare a Skill Set for this project, but do not apply it yet.”

The installed `SKILL.md` is the operating contract. Customer docs explain the
goal and decision points; the skill contains version-matched command routing.

## Verify

Ask the agent to report the SelfTune version, connected harnesses, discovered
skills, evidence coverage, changes made, and unresolved decisions. Operational
work is incomplete if it only reports that a command exited successfully.

Use the installed SelfTune skill as the canonical procedure. Translate user
intent into the smallest applicable workflow. Inspect first, state mutations,
verify outcomes, and stop for consent before hooks, history import, background
services, remote connections, publishing, credentials, billing, or deletion.