Форензический аудит научных статей

★ 7.1 · testing

integrity-forensics is a Claude Code skill that runs a forensic integrity sweep on scientific papers by acting as a thin launcher for the Anti-Autoresearch project, covering 46 detection patterns across 8 families — including a deterministic GRIM/GRIMMER/statcheck core with span-anchored evidence. The skill clones the upstream repository at a fixed SHA commit, validates the pin through an eval gate testing 8 injected-defect classes, then delegates execution unchanged with no reviewer knobs exposed. After the sweep completes, the upstream verdict (CLEAN_GIVEN_EVIDENCE / SOFT_FLAGS / HARD_FLAGS / REVIEW) is post-processed into a typed policy gate (BLOCK/WARN/NO_NEW_BLOCKER) and an append-only obligations ledger. It accepts a paper directory, PDF, or arxiv-id as input and is triggered by phrases like "integrity forensics" or "forensic audit this paper", as well as automatically from /paper-writing, /peer-review, and /resubmit-pipeline.