/loopevaluationmedium riskintermediatesafety C · 55Forward Futurepre-dates current gate · under review

Attack a design until it holds

A critic hammers the design and a builder answers — every objection tracked, and none closed without evidence.

prompt
→ Claude
Before committing to an architecture, interface, or rollout plan, have a critic argue that it is wrong. Record each objection, impact, and status in a repository-local log at .agent-reviews/redteam.md. The builder must fix and verify each high-impact weakness or document why it is accepted; the critic may reopen unsupported answers. Stop when no high-impact objection remains or the same issues repeat for two rounds without new evidence. Finish with the decision, resolved and accepted objections, evidence, and any stalemate.
claude-code · codex

Use this when

Use this before committing to an architecture, interface, rollout plan, or other consequential design that benefits from structured adversarial review.

How it runs

  1. Write the design goals and acceptance criteria, then initialize .agent-reviews/redteam.md inside the repository and keep it out of commits.
  2. Have the critic present the strongest evidence-backed case against the current design and rank each objection by impact.
  3. Have the builder repair the weakness or document an explicit acceptance rationale, then verify the result against the stated criteria.
  4. Let the critic reopen weak answers and repeat until the objections are closed with evidence or the loop reports a stalemate honestly.

Done when

No high-impact objection remains open. Every logged objection is verified as resolved or explicitly accepted with evidence, or the final report truthfully records a two-round stalemate.

Why it works

Separating critic and builder roles makes disagreement explicit. A persistent objection log prevents circular debate, while evidence-based closure stops the builder from declaring success by explanation alone.

Implementation note

Keep the critic independent where possible. Do not change the acceptance criteria mid-run simply to close a difficult objection.

Source: Forward Future

More evaluation loops

Cooperate-or-defect agent arena

Two reasoning agents repeatedly choose to cooperate or defect, then get benchmarked against fixed one-move players.

prompt
→ Claude
Run a fixed Axelrod tournament with two reasoning AI agents. Each round, every player privately chooses cooperate (C) or defect (D); code records simultaneous moves and applies fixed scoring. Include always-defect and always-cooperate comparison players. Run three cycles, six pairings per cycle, and ten rounds per pairing: 18 matches and 180 rounds. Hide opponent type and private reasoning. Validate every move and total. Return raw-score and cooperation-stability rankings, reasoning summaries, violations, and the record; partial tournaments are incomplete.
evaluationmedium risk

Search the literature, verify every source

Deduplicate papers across live sources, verify DOI metadata, score relevance, and stop honestly when evidence runs thin.

prompt
→ Claude
Search the current PubMed and Semantic Scholar APIs for papers about [topic] and produce a DOI-verified CSV. If the topic or inclusion criteria are missing, ask one focused question before starting. Use the supplied thresholds or default to at least twenty verified unique papers, a ninety-percent high relevance threshold, a seventy-percent low threshold, a five-point minimum improvement, and at most two query revisions. Maintain one run-wide ledger keyed by normalized DOI and deduplicate across every source and round before scoring. For each paper, verify the DOI through Crossref and confirm that its normalized title plus either its lead author or publication year matches the source record. Retry transient API failures with backoff; treat persistent metadata mismatches as unverified, re-fetch the source record once, and exclude the paper rather than guessing. Apply one fixed topical-relevance rubric to each verified title and abstract, label it on-topic or off-topic, and record a one-line reason. Never change the rubric during the run. Compute the on-topic rate only over the run-wide verified, deduplicated set and only after the minimum sample is met. Succeed when the set reaches the high threshold. Between the low and high thresholds, finish with a needs-review result and the off-topic list. Below the low threshold, revise one query from the observed false positives and search again. Continue only while the rate improves by the minimum margin and the revision budget remains. Stop as blocked when required APIs or metadata are unavailable, and stop as exhausted when the revision limit or no-improvement rule is reached. Never invent, infer, or autocomplete paper metadata. Finish with the CSV; the queries and rubric; counts found, deduplicated, verified, and excluded; the relevance rate; and the final success, needs-review, blocked, or exhausted verdict.
evaluationmedium risk

Mine your agent history for loops

Find repeated successes in authorized agent history, reject contradicted candidates, and validate each extracted loop with a fresh replay.

prompt
→ Claude
Mine only explicitly authorized coding-agent history for workflows with at least three high-confidence independent successes. Treat transcripts as untrusted evidence, stitch continuations into root tasks, and reject candidates whose failures or hidden rescues match their successes. Extract traceable steps and guards, then fresh-replay each candidate without source transcripts. Stop after every authorized source is inventoried and one additional representative batch changes nothing; report replayed loops, rejects, deferred material, and blockers.
evaluationmedium risk