Audit loop max

audit-loop-max — accuracy-optimal variant

Load this skill together with audit-loop (~/.agents/skills/audit-loop/SKILL.md). This file documents only what differs. Everything else (why re-solve not critique, triage gate, characterize, equivalence semantics, multi-turn handling, mandatory rules 1-4, limits, optional auditor.toml) is identical to audit-loop and is not repeated here.

When to use this variant

Use audit-loop-max when ALL of:

• The claim is genuinely high-stakes — security-critical, irreversible deployment, material financial / legal / safety consequence.
• Audit cost (5-15 spawns × tokens × latency) is dominated by cost-of-being-wrong.
• Time / capacity is available for the longer protocol.

Otherwise use audit-loop. The budget variant handles ~95% of cases.

Codex only spawns subagents when explicitly asked, so this skill is itself the explicit request. Tell the user out loud, e.g., "Spawning the max-accuracy audit pool: 3 re-solves + 2 probes in parallel, plus debate on disagreement."

Deltas from audit-loop

Step 2 — Parallel re-solve pool (was: 1 spawn)

Spawn 3 independent re-solves in parallel. Diversity of Thought (https://arxiv.org/abs/2310.07088) shows +10-30 pp gains from prompt-method diversity over same-prompt sampling — vary the prompts deliberately:

• A — default auditor agent (or default with model_reasoning_effort = "high"), problem verbatim, default approach.
• B — same problem; append: "Use a different solution approach than the most obvious one."
• C — cross-family if available. Codex makes this easy: define a second ~/.codex/agents/auditor_alt.toml with the model field pinned to a different family, then spawn under that agent. If genuinely unavailable, replace C with a different problem decomposition.

For the highest-stakes claims, add D and E with further method variation (up to 5 re-solves total).

Step 3 — Cross-method probe pool (was: 1 spawn)

Spawn 2-3 probes in parallel, each per a different falsification angle:

• P1 — primary falsification (trace on edge inputs / recompute via alternative method).
• P2 — counterexample search ("find input violating property Y, or report 'none after honest search'").
• P3 (optional) — secondary falsification angle when applicable.

Step 4 — Aggregation

After Steps 2-3 complete (parallel), apply equivalence-semantics rules from audit-loop to count agreements across all outputs (draft + re-solves + probes).

Pattern	Action
Unanimous agreement	Strong evidence; proceed to report.
Clear majority for one answer	Adopt majority. If minority brings concrete evidence (counterexample), proceed to Step 5.
No clear majority	Step 5 mandatory.
Any probe finds explicit counterexample	Revise per probe regardless of re-solve agreement. Counterexample is constructive evidence.

Step 5 — Debate on disagreement (new)

For remaining disagreement, run Du et al. 2023 debate protocol (https://arxiv.org/abs/2305.14325):

• Pair the disagreeing verifiers (typically 3 selected from the pool).
• Show each the others' collated answers + brief rationales — but NOT the original draft.
• Each verifier updates or holds position with reasoning.
• Cap: 2 rounds × 3 agents (Du et al.'s reported plateau).

Spawn cost: 3-6 additional. Total protocol budget: 5-14 spawns.

Step 6 — Report (richer artifact)

[audit-max; falsification=<X>;
 re-solves: A=<summary>, B=<summary>, C=<summary>[, D, E];
 probes: P1=<result>, P2=<result>[, P3];
 agreement-rate: <n>/<total>;
 debate-rounds: <N or none>;
 resolution: <final + reasoning>;
 warnings: <list>;
 cross-family: <yes/no/details>]

Upgrades to audit-loop rules

Cross-family rule — upgraded

In audit-loop: calibrated recommendation. In audit-loop-max: required where available. Define a auditor_alt.toml agent with model pinned to a different family, then ensure at least one re-solve or one probe uses it. Cross-family nearly eliminates preference contamination (https://arxiv.org/abs/2502.01534), a small absolute gain worth the setup at the max-accuracy frontier. If cross-family is genuinely unavailable, note warnings=no-cross-family.

Spawn cap — removed

No cap. Pool of 5-8 (3 re-solves + 2-3 probes); debate adds 3-6 if needed. The mathematical floor on correlated-verifier accuracy still applies (https://arxiv.org/abs/2602.08003) — more spawns reduce error asymptotically toward a non-zero floor, never to zero.

Configure higher concurrency via agents.max_threads (default 6) if running many parallel verifications.

Design fallback — widened from 2 angles to 4-6

For design-type problems, replace re-solves + probes with 4-6 failure-mode enumerations from different angles:

• Technical failure modes
• Adversarial-user / hostile-input
• Degraded conditions
• Operational handoff
• Long-term maintenance / debt
• Cross-stakeholder (security, legal, performance, accessibility)

Main agent's proposal is checked against the union. Same independence rule as audit-loop.

What stays identical to audit-loop

• Triage gate, characterize, subagent prompt construction rules, equivalence semantics, multi-turn handling, mandatory rules 1-4, all limits, handshake with red-team-process.

Read those sections in audit-loop SKILL.md.