code-verifierlisted

# /code-verifier Audit the GENUINENESS of evidence — not just that a command ran. ## Master TOC - [Core principle](#core-principle) - [The Three-Layer Gate](#the-three-layer-gate) - [Fake-Run Patterns](#fake-run-patterns) - [A. Tests that pass without testing](#a-tests-that-pass-without-testing) - [B. Scripts that fabricate results](#b-scripts-that-fabricate-results) - [C. Pipelines that short-circuit](#c-pipelines-that-short-circuit) - [D. ML-research specific](#d-ml-research-specific) - [When to invoke](#when-to-invoke) - [Output format when fake pattern found](#output-format-when-fake-pattern-found) - [Anti-rationalisation](#anti-rationalisation) - [Companion](#companion) ## Core principle **"Ran a command" ≠ "verified the work."** A test that always passes (because it asserts `True`) proves nothing. A script that prints `mAP = 0.85` from a hardcoded literal proves nothing. A pipeline that swallows exceptions silently and returns a default proves nothing. This skill audits the GENUINENESS of evidence — not just that a command ran. ## The Three-Layer Gate Before any completion claim involving code / tests / results, run all three layers in order: 1. **Layer 1 — Did it really run?** Defer to `superpowers:verification-before-completion`. Run the exact command in this turn; read full output; check exit code. Skip if already cleared in this turn. 2. **Layer 2 — Is the run genuine?** Audit the artifact for the FAKE-RUN patterns below. If any match, the run is NO