Agent skill

Command examples

Commands use stim. If it is not installed globally, replace stim with npx stim.

Install the bundled skill from the repository:

npx skills add appandflow/stim

In a terminal, the skills CLI asks which agents to install to when it cannot tell. Without a terminal, as in a script or Stim Desktop's setup guide, add --yes: it installs to the agents it detects, or to ~/.agents/skills when it detects none.

npx skills add appandflow/stim --yes

The installed skill is named stim. It is a small discovery router that asks the agent to load stim guide agent before using Stim, or npx stim guide agent when stim is not on PATH. The normal workflow, ownership rules, destructive-command rules, and routing to detailed topics all come from the installed CLI and therefore match its version.

You normally ask for the outcome: "build and run the app on iOS", "show the recent app errors", or "run this on my connected phone". Those requests match the skill without naming Stim. Add "use Stim" only when you want to override a project wrapper or another tool choice.

An agent works in a separate worktree by default. Ask it to work in the current checkout when you want that; it also does so when the task depends on your uncommitted changes there, and says so.

Upgrading Stim also upgrades the guidance. The static skill does not need to be reinstalled when commands or behavior change.

Stim handles local build, install, launch, and readiness checks itself. The skill does not require a device automation package. An agent can use one separately when a task needs taps, text input, snapshots, screenshots, or recordings.

Clean up when a session ends

An agent that skips stim stop leaves a Metro server and a device running. For Claude Code, add a SessionEnd hook that stops the workspace in the session's own working directory when the session ends:

settings.json
{
  "hooks": {
    "SessionEnd": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "node -e 'let d=\"\";process.stdin.on(\"data\",c=>d+=c);process.stdin.on(\"end\",()=>require(\"child_process\").spawnSync(\"npx\",[\"stim\",\"stop\"],{cwd:JSON.parse(d).cwd,stdio:\"inherit\"}))'"
          }
        ]
      }
    ]
  }
}

Claude Code passes the hook a JSON payload on stdin whose cwd field is the session's working directory; the command above reads it and runs npx stim stop there, since a hook cannot assume stim is installed globally. Codex CLI has no equivalent session-end hook today (see openai/codex#20374).

Clean up when a session ends​

Clean up when a session ends