> ## 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.

# CLI reference

> Find the stable intent-level SelfTune commands and the version-matched source for exact flags.

The CLI is SelfTune's operational API. Start with the intent-level commands
below. Use the installed command's `--help` output as the source of truth for
exact flags because the agent skill and CLI ship together.

## Run a command

```bash theme={null}
selftune <command> [options]
```

Without an executable on `PATH`, use:

```bash theme={null}
npx selftune@latest <command> [options]
```

## Customer command map

| Intent                         | Command                                          | Detailed reference                |
| ------------------------------ | ------------------------------------------------ | --------------------------------- |
| Check installation             | `selftune doctor`                                | [doctor](/cli/doctor)             |
| Import approved local evidence | `selftune sync`                                  | [sync](/cli/sync)                 |
| Inspect health                 | `selftune status`                                | [status](/cli/status)             |
| Open the local UI              | `selftune dashboard`                             | [dashboard](/cli/dashboard)       |
| Preview an improvement         | `selftune improve --skill-path <path> --dry-run` | [improve](/cli/improve)           |
| Verify a draft                 | `selftune verify --skill-path <path>`            | [create and verify](/cli/create)  |
| Publish an approved draft      | `selftune publish --skill-path <path>`           | [create and publish](/cli/create) |
| Manage project Skill Sets      | `selftune sets <subcommand>`                     | [sets](/cli/sets)                 |
| Manage the local service       | `selftune service <subcommand>`                  | [service](/cli/service)           |

Low-level grading, evaluation, evolution, search, and orchestration commands
remain available for debugging and expert control. They are not required for
the normal customer journey.

## Machine-readable output

Use `--json` when the selected command supports it. Consumers must still check
the returned state and target revision; a zero exit code alone does not prove
that a skill improved or a service is healthy.

## Safety boundaries

Local inspection, skill mutation, background operation, authentication, remote
upload, publication, and deletion are distinct actions. Preview mutations and
request user approval at the boundary that matters.

See [CLI and headless use](/run/cli) for automation guidance and
[Privacy and permissions](/guides/privacy-and-permissions) for data boundaries.

Prefer the intent-level command map. Before executing a mutating or remote
command, inspect `selftune <command> --help` for the installed version and state
the expected changes. Use structured output where available and verify the
resulting state, artifacts, and target revision. Do not fall back to a
low-level command merely because it exposes more flags.