> ## 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.

# Set up with your agent

> Give your agent a bounded prompt to install SelfTune and verify the first local result.

SelfTune is designed to be operated by an agent. Give the agent the goal and
the consent boundaries; let it discover the platform-specific commands.

## Copy this prompt

```text theme={null}
Set up SelfTune in this environment and verify that it works.

First inspect the operating system, available Node.js or Bun runtime, current
agent harness, installed skills, and whether SelfTune is already installed.
Choose the agent-skill path for interactive use unless this is a headless
environment. Explain your choice before making changes.

Install SelfTune from selftune-dev/selftune, initialize it for the supported
agent harnesses present here, and run its health check. Before installing
hooks, importing historical sessions, or enabling a background service, tell
me exactly what will change and ask for approval.

Keep this setup local. Do not create or connect a cloud account, upload data,
publish a skill, change credentials, enable billing, or delete existing data.

After setup, sync approved local evidence and show me the skill inventory.
Report:
- the installation path and SelfTune version;
- which agent harnesses are connected;
- which skills were discovered;
- which skills have evidence and which have an honest no-data state;
- any warnings or actions that still need my decision.

Stop when the local health check and inventory are verified. Do not start an
improvement or modify a skill yet.
```

## What the agent should do

<Steps>
  <Step title="Inspect before installing">
    The agent checks the runtime, operating system, harness, installed skills,
    and existing SelfTune state. It should reuse a healthy installation rather
    than layering another one over it.
  </Step>

  <Step title="Explain changes and request consent">
    Installing the skill is expected. Hooks, historical imports, background
    services, remote connections, and credentials have separate consequences
    and must be disclosed.
  </Step>

  <Step title="Install and initialize">
    For an interactive agent environment, the normal install is:

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

    The agent then follows the installed SelfTune skill's initialization
    workflow for the harnesses it found.
  </Step>

  <Step title="Verify, do not assume">
    The agent runs the SelfTune health check, imports only approved local
    evidence, and inspects the resulting inventory. It reports no-data states
    honestly rather than manufacturing a health conclusion.
  </Step>
</Steps>

## When to restart your agent

Some harnesses load their available skills only when a session starts. If the
agent cannot see SelfTune immediately after installation, start a new session
and paste this shorter prompt:

```text theme={null}
Continue the SelfTune setup. Inspect the existing installation before changing
anything, then finish the local health check and skill inventory. Preserve the
same consent boundaries as before.
```

## Expected result

The agent should finish with a concise report, not a stream of command output.
The report must identify the installed version, connected harnesses, discovered
skills, evidence coverage, blocking errors, and deferred decisions.

Continue to [Get your first result](/quickstart) if you want to perform the
manual verification yourself.

Treat the copyable prompt as the controlling task contract. The installed
SelfTune skill is the canonical source for platform-specific operating
procedures. Prefer read-only inspection before mutation. Keep a record of each
change made. If installation instructions conflict with the detected version,
run the current package's help or doctor command and report the discrepancy
instead of guessing. Stop for explicit approval at every consent boundary
listed in the prompt.