security-hardening-review-opslisted

# Security-hardening review operations How the PR #83 skill-vetting campaign was run — reviewer orchestration and delivery-governance lessons, most surfaced by the maintainer steering the process. These are **project observations and applications**, not replacements for the installed `cross-model-review`, `delegation-and-review`, and `operational-rigor` doctrine; the stop-condition and model-selection rules defer to those (OPS-1/3/4/5/7/12 cite a canonical rule), while OPS-2/6/8/9/10/11 are marked project observations with no canonical counterpart. Evidence note: unlike the other three skills, several OPS incidents are **history-only, sourced from the session transcript and the gitignored `internal/gate-b-2026-07-25/` ledger — NOT independently repo-verifiable**; the canonical-rule citations (to the installed skills) and the `reviews/2026-07-25-skill-vetting-*.md` design records DO resolve. ## Reviewer orchestration ### OPS-1 — one pass under-samples; run ≥2 blind passes per family - **Trigger:** treating a single reviewer pass (even a strong model at max effort) as full coverage of a family's lens. - **Do:** run at least two independent, mutually-blind passes per reviewing family. A **pass** = one reviewer result; a **campaign round** = one packet → fix cycle. A family's lens is "clean" only after two CONSECUTIVE passes surface nothing new (delegation-and-review §3's miss-costly-audit rule — one clean pass is not convergence); reset the counter whenever a pass f