referee-panellisted

# /referee-panel — pre-submission mechanical auditor panel A multi-agent **finding-level** red-team of a paper build. Each of ~12 lenses audits ONE failure mode, returns structured findings under a shared `cannot_verify`-or-cite contract, every finding is then **adversarially verified** (a skeptic tries to refute it), and an editor **dedups, ranks, and writes a memo**. Pattern borrowed from `github.com/Ingar30/reviewer`. This is the **mechanical-failure-mode** axis of review. It complements, and does not replace: - **community-persona referee reports** (`docs/reference/referee_reports/R*.md`) — the "which reviewer community reacts, and how" axis; - **`/findings-audit`** — stress-tests headline findings against *external* evidence; - **`/style-review`** — prose. Output is AI-generated: **triage before acting**. The memo's "Cannot verify" section names what artifact would settle each open question. ## Arguments - `/referee-panel` — infer the project from cwd; audit its primary maintained build (see "Resolving the build" below). Confirm the build with the user if ambiguous. - `/referee-panel <build>` — audit a specific build, e.g. `/referee-panel paper/oe.tex` or `/referee-panel projects/deterrence paper/science.tex`. - `/referee-panel --lenses id,claim-evidence,numerical,power-multiple-testing` — run only a subset of lenses (keys below). Default runs all 12. ## What this skill does (the flow) 1. **Resolve the build and its sources.** Identify the driver `.tex