Playground¶

The tool room, live in your browser. The editor below is a tasks.py whose tool calls all go through toolroom's typed handles; the prompt is fm — footman, toolroom's host, which the handles detect and route through on every call. Python (and both packages) load into the page via Pyodide on first use — nothing is installed on your machine, and nothing you type leaves it.

The browser sandbox, said plainly: it has no processes and one thread, so every spawned child is simulated — it succeeds and its output says [simulated] — and runs are sequential. Everything else is the real thing: the flag translation, .argv building whole command lines without running anything, taught errors, --json, --dry-run plans — and fm test runs the real pytest, in-process through the handles, right here in the page. The prompt completes too: press Tab and the candidates come from footman's manifest walk, rebuilt from whatever the editor says.

Press Run. The gate fails — one of the tests is wrong on purpose. Read pytest's diff, fix fizzbuzz (or the test), and run it green. Then try fm ship to watch .argv build a docker compose line without a docker in sight, --dry-run lint --fix to see the exact command a flag becomes, and deploy produ to read a taught error.