setuplisted

# vteam setup — proof-of-done for AI agents First explain, in substance, these 5 lines and nothing grander: 1. Agents say "done"; vteam makes them prove it — machine gates that actually go red. 2. Every checking gate ships a `--selftest` mutation proof (green fixture + red mutations): the gate itself is tested, not trusted. 3. Evidence is pinned to commits — claims carry command + real output written to files, not chat scrollback. 4. `.vteam/manifest.json` guards updates (user edits are never silently clobbered) and the secret scan fails closed. 5. The same gates drive Claude Code, Cursor, Windsurf, Codex, and Copilot — one team, any agent tool. Then run the funnel in order. Show the user each command's REAL output; never claim a step succeeded without it. ## 1. Grade the repo first ``` npx vteam-harness audit ``` Scores the repo 0-100 on proof-of-done before changing anything, so the user sees the gap the install closes. If the installed vteam-harness predates the audit command (the CLI prints usage without `audit`), say exactly that and continue with step 2 — do not fake a score. ## 2. Install ``` npx vteam-harness init ``` Interactive; `--yes` accepts defaults, and flags like `--key PROJ --profile node --tools claude-code` skip prompts. Ask before running it: it writes `vteam.config.yaml`, `.vteam/` (gates + manifest), `.claude/` (skills + SessionStart hook), `.githooks/pre-push`, CI workflow, and docs skeletons. It refuses to run twice — an existing install updat