What problem does it solve? LLM agents can unintentionally produce fraudulent experimental results, such as fake ground truth derived from model outputs, self-normalized scores, phantom results referencing nonexistent files, and overstated evaluation scope. This Skill adds an independent integrity constraint by having an external reviewer model audit the experiment before claims are written. ## Core Features & Use Cases - Cross-Model Integrity Review: The executor only collects file paths while an external reviewer backend (Codex MCP or Manual Review MCP) reads the code and judges integrity, preserving reviewer independence. - Six-Point Audit Checklist: Checks ground truth provenance, score normalization, result file existence, dead code detection, scope assessment, and evaluation type classification, each with PASS/WARN/FAIL verdicts and file:line evidence. - Structured Reports: Writes EXPERIMENT_AUDIT.md and EXPERIMENT_AUDIT.json with an overall verdict, per-check details, action items, and claim impact tags consumed by downstream pipeline skills. - Use Case: After an autonomous research pipeline finishes experiments, run the audit to verify that reported metrics come from real dataset ground truth and that every claimed number matches an actual result file before writing the paper. ## Quick Start Ask the agent to audit the experiment results in your project directory for integrity before writing any claims.
