Audit every command in README runs as written

[bordeauxred]

Why

Broken fenced commands are one of the fastest ways to lose trust with engineers who land from a newsletter. A single `pip install ... && some-command` that exits non-zero and the reader closes the tab. The doc-command tests in #29 will eventually automate this; in the meantime, do one manual pass.

What

• Walk every fenced `bash` block in `README.md` top-to-bottom and run it on a fresh machine (or fresh venv + fresh clone).
• For each block, record: command, exit code, first 5 lines of output.
• Fix any divergence:
  • command that fails → fix command or remove from README
  • output that doesn't match what's shown → update sample output
  • command that requires undocumented setup → document the prerequisite
• Repeat the same audit for any `bash` blocks in the `docs/` site that an external visitor would reach via README links.
• Repeat for `CONTRIBUTING.md` and `docs/getting-started.md`.

Acceptance

Every fenced command in README runs to completion on a fresh `pip install clawloop` venv (allowing for the documented optional steps). No surprises.

Why this is launch-blocking

Codex review (2026-04-19) flagged the absence of a manual audit as a missing P0. Doc-command parsing tests (#29) are the long-term fix; this is the launch-day backstop.