# Check OK — portable honesty check (any AI tool) **Not Cursor-only.** Use this in Claude Code, GitHub Copilot, ChatGPT, or any assistant. ## Fast path 1. In the repo terminal: ```bash ok check-ok --topic "" --scaffold-only ``` 2. Fill the scope in the created `docs/reviews/-.md` file. 3. Paste the block below into your AI chat (or type **Check OK** / `/check-ok` in Cursor / Claude Code if skills are installed). 4. Mechanical gate: ```bash ok check-ok --path docs/reviews/.md ``` The CLI line is **mechanical only** — it writes `gate: mechanical` and **cannot authorize Auto**. Auto authorization requires a digest-bound `freeze_review` ledger entry (`ok ledger append --kind freeze_review`). An explicit `auto_may_start: false` at the freeze-block top level blocks Auto regardless of ledger or stamp; the CLI never writes or clears that key. Author freeze-review-loop `pass` is also **not** Auto-cleared when `honesty.adversarial_freeze` is `suggest` or `require` — see `docs/ADVERSARIAL-FREEZE-REVIEW.md`. ## Paste-ready prompt ``` Check OK — run the Overseer honesty gates on my current work. Model: thinking-high for review rounds 1) If no freeze artifact exists, scaffold with: ok check-ok --topic "" --scaffold-only then fill Scope + seven-tier matrix in that doc. 2) Freeze review (beginning gate): follow .cursor/skills/freeze-review/SKILL.md or .claude/skills/freeze-review/SKILL.md — cite every finding as path:line. Mechanical: ok check-ok --path 3) If code already landed: build-verification-review against the freeze + git diff. 4) Verdict: pass | findings | blocked. Do not invent a new docs.lanes entry. 5) Escalate to human on security / irreversible / real_money / gates_tier3. Read: docs/CHECK-OK.md; docs/OVERSEER-KIT-SPEC.md §6. ``` ## Runtime map | Tool | Skill auto-load | Fallback | | --- | --- | --- | | Cursor | `.cursor/skills/check-ok/` after `ok sync` | CLI + this paste | | Claude Code | `.claude/skills/check-ok/` after `ok sync` | CLI + this paste | | Copilot / others | — | CLI + this paste (skills not native) |