← ClaudeAtlas

light-auditlisted

Run the dual-audit panel at its shortest depth — one independent round instead of three. Use for a small disagreement, a suspected mistake, or a question a deterministic run did not settle, and whenever the user asks for a light review, a quick second opinion, or "check this with the other model". Same panel, same gates, fewer rounds.
ttomasyoung/dual-audit · ★ 1 · AI & Automation · score 69
Install: claude install-skill ttomasyoung/dual-audit
# Light audit (the panel, one round) This is the **same panel** as `dual-audit`, run at its shortest depth: `mode: 'quick'`, one round instead of three. Same seats, same independent first-round reading, same gates, same terminal states. You give up cross-examination and nothing else. It is worth knowing what that replaced, because it changes what you can use this for. The old cheap lane was a different mechanism: a single read-only reviewer, dispatched with a brief *you* had written — so it read your framing of the problem instead of the problem, which is the exact failure this project exists to prevent — and with no Claude-side seat, so a question about whether code *works* was answered by reading it. Quick depth keeps the independent reading and the run seat. ## When this is the right depth - a small disagreement or a suspected mistake, and one round of independent reading settles it; - a deterministic run was attempted and did not settle the question; - the user asked for a quick second opinion; - closing out individual points a full panel left unresolved. Go to full depth instead when the question touches what the panel exists for: a definition, threshold or rule others will rely on; output that becomes a load-bearing input downstream; rules, schemas, permissions or critical configuration; an irreversible, destructive or production action; or anything in `critical_areas` in `__DUAL_AUDIT_PROFILE_PATH__`. **One round is shorter, never softer.** The gates are identica