checklisted

# Are this repository's checks real? A check that cannot fail looks exactly like a check that passes. `|| true`, `continue-on-error`, a linter pointed at an empty directory, a test with no assertion — all of them print a green tick. This skill answers the only question that separates them: **can it go red?** ## What to run ```bash npx -y agent-quality-kit doctor --run ``` Read the verdict carefully; the exit code alone is not the answer. **Three outcomes per gate, and they are not two.** Never collapse them: | what you see | what it means | what to do | | --- | --- | --- | | `✔` | ran, and found nothing | nothing | | `✘ … код N` / `exit N` | a finding about the code | fix the code | | `не смогли проверить` / `could not check` | **the check itself failed** — tool missing, unexpected exit code, killed | fix the tooling; do not touch the code it named | The third one is the trap. If you report it as a finding, the human will go and "fix" a file that was never broken, and the broken instrument stays broken. ## What else is worth running - `npx -y agent-quality-kit probe` — plants a known-bad sample into a **copy** of the file the repository's own history shows as the most repaired, then runs the declared gates against it. A class nobody catches is the most valuable thing this tool produces. It never touches the working tree and always exits 0. - `npx -y agent-quality-kit prompt` — turns everything above into one ordered task list where **every item carries the co