ai_stp

# Quickstart ## Install the CLI The CLI installs as an ordinary `uv` tool: ```bash uv tool install ai-stp-cli ``` Then check that the command is available: ```bash ai-stp version ai-stp doctor --json ``` `doctor` reports the state of the environment, what this installation can do, and anything that has to be fixed before assembling a setup. ## Prepare the device For local work without an account, create a developer passport and a device: ```bash ai-stp passport developer init --json ai-stp device init --json ``` The device gets a `device_id` and a key. The private key is held in the operating system's secret store; where that is unavailable, the CLI says plainly that it fell back to file storage. ## Look at the catalog Anonymous reading of the public catalog needs no sign-in: ```bash ai-stp registry search --json ``` For one exact object, use `show`. With no network, the CLI may answer from the last confirmed local copy and will say when the platform last confirmed it. ## What comes next After the first `doctor`, the agent reads the machine help: ```bash ai-stp help --agent --json ``` That help is the source of commands for the Agent Skill. An agent should not invent commands from memory when the CLI already answers with them. === "I am choosing a published setup" Open the [catalog](catalog/index.md) and check the trust line, the compatibility with your [harness](harnesses.md), and what the setup is made of. === "I am assembling my own setup" Start with [components](components/index.md): it helps decide what belongs in a `skill`, what in an `mcp`, and what stays a `setting` or an `instruction`. === "I am publishing a component" See [publishing](publishing/index.md) and the trust rules. A published version is immutable, so it is worth checking the passport before the first release.