Create a factory verification skill

Write .factory/skills/verify-<repo>/ for the next agent, not as a human tutorial. Interview the repo before asking the user.

Interview

Answer these from the checkout:

• Surface: what a user touches (web UI, CLI, API, library).
• Run: the repo's own dev or start command, port, env, and seed data.
• Drive: an existing harness first, then Surf for a web UI or the real CLI for a terminal tool.
• Observe: screenshots, transcripts, response bodies, logs, exit codes.
• Isolate: a second instance must not reuse the user's port, database, or Surf session.

Then generate the files:

node <factory-supervise>/scripts/write-verification-skill.mjs \
  --code-root <code-root> \
  --factory-root <factory-root> \
  [--repo <repo-name>]

Replace the generated placeholders with commands and selectors you actually observed. The skill needs Launch, Doctor, Drive, Evidence, and Cleanup. Evidence goes to .factory/evidence/<ticket>/<task>/ and must survive cleanup.

For a web UI, the Drive section must use a unique SURF_SESSION, a worktree-owned free port, surf session.ensure, surf go, surf wait.ready, surf read, desktop and narrow screenshots, surf console, surf network.stats, surf smoke --screenshot, and --auto-capture on failure.

Prove one feature

Run the generated skill once: launch, doctor, drive one feature, capture evidence, clean up, then confirm the evidence file is still there. Fix the skill if that fails. A skill that was never executed is a draft.

Store a rerunnable command in FACTORY.json as verification.command when one command can regenerate the manifest. The lifecycle hook still requires the manifest; it does not trust the command name.