CLI
How a user and an agent work with the ai_stp CLI.
CLI
The CLI is the main user interface of ai_stp for assembling, checking and
installing setups. The web shows the catalog and the account; it does not apply
configuration to a harness.
Commands that observe
Observing commands create nothing and change nothing:
ai-stp version
ai-stp doctor --json
ai-stp capabilities --json
ai-stp help --agent --json
ai-stp config show --json
ai-stp auth status --json
ai-stp device show --json
On a fresh installation they return a typed state rather than quietly initialising a device.
Commands that change
Changing commands say what they do:
ai-stp device init --json
ai-stp passport developer init --json
ai-stp passport developer update --json
ai-stp passport device refresh --json
Dangerous actions ask for confirmation. Resetting a device, for one, must not happen as a side effect of a diagnostic.
Install telemetry
Off by default. After explicit consent the CLI sends one anonymous request per component actually installed — what it carries and why is in Install telemetry.
ai-stp telemetry show --json
Agent Skill
The Agent Skill starts with:
ai-stp doctor --json
ai-stp help --agent --json
After that the agent uses the machine list of commands from the CLI itself. That is what protects against stale instructions and hallucinated flags.
| Stage | Command | Why |
|---|---|---|
| Check the environment | ai-stp doctor --json | see what is available and what is broken |
| Learn the capabilities | ai-stp capabilities --json | do not assume harness support |
| Get the machine help | ai-stp help --agent --json | use the commands that exist now |
| Find an object | ai-stp registry search --json | choose a setup or a component |
| Read an object | ai-stp registry show <stable_id> --json | check the passport before deciding |
!!! note "Commands here are a guide for people" If the CLI answers with newer machine help, the agent follows that. This documentation explains the model; it does not replace the CLI's executable help.