Run

Run the app

Goal: actually launch this project and exercise the change, so "it works" means observed behavior — not a green test suite. $ARGUMENTS may name what to run or which change to exercise; if empty, infer from the recent work.

1. Find how this project is meant to run (in priority order)

1. A project skill/command that already launches it — check .chainlesschain/, .claude/, package.json scripts (dev, start, serve), Makefile, Procfile, docker-compose.yml, README "Quick Start".
2. The dev command for the detected stack. Prefer the one a human would use locally, not the production/CI one.

2. Launch pattern by project type

• CLI tool: run the binary with a realistic subcommand + flags (e.g. node bin/<cli>.js <cmd> --help, then a real invocation). Read stdout/stderr.
• HTTP/API server: start it in the background, poll the health/route until it responds, hit the relevant endpoint with curl, then stop it. Never leave a server running — capture its task id and kill it when done.
• TUI: launch with a scripted/non-interactive path if one exists; otherwise describe the exact keystrokes and capture the first screen.
• Electron / desktop: use the project's dev script (e.g. npm run dev); if there's no display, say so and fall back to building + smoke-importing the changed module.
• Browser/web app: start the dev server, open the route that exercises the change, take a screenshot or read the rendered DOM/console.
• Library (no app): there's nothing to "launch" — write a tiny throwaway script that imports the changed API and runs the new path, then run it.

3. Drive the change, then report

• Exercise the specific path the change touched (the new flag, route, screen, function) with realistic input.
• Report what you OBSERVED: the actual output / status code / screenshot / log line — quoting it. Distinguish "ran and behaved correctly" from "started but I couldn't reach the changed path".
• Clean up: stop any background process you started. If you couldn't run it (no display, missing service, needs a device), say exactly what's blocking and what you'd run once unblocked — don't claim success you didn't observe.