Auto Review Loop: Autonomous Research Improvement

Autonomously iterate: review → implement fixes → re-review, until the external reviewer gives a positive assessment or MAX_ROUNDS is reached.

Context: $ARGUMENTS

Constants

• MAX_ROUNDS = 4
• POSITIVE_THRESHOLD: score >= 6/10 AND verdict ∈ {"ready", "almost"} — both must hold, matching the operative STOP CONDITION below. Verdict vocabulary is {"ready", "almost", "not ready"}. (Earlier wording used "or" + a stale verdict set; the AND form is authoritative.)
• REVIEW_DOC: review-stage/AUTO_REVIEW.md (cumulative log) (fall back to ./AUTO_REVIEW.md for legacy projects)
• OUTPUT_DIR = review-stage/ — All review-stage outputs go here. Create the directory if it doesn't exist.
• REVIEWER_MODEL = gpt-5.5 — Model used via a secondary Codex agent. Must be an OpenAI model (e.g., gpt-5.5, o3, gpt-4o)
• REVIEWER_BACKEND = codex — Default: Codex reviewer agent at xhigh reasoning. Override with --reviewer: oracle-pro only when the user explicitly requests Oracle; if Oracle is unavailable, warn and fall back to Codex xhigh.
• HUMAN_CHECKPOINT = false — When true, pause after each round's review (Phase B) and present the score + weaknesses to the user. Wait for user input before proceeding to Phase C. The user can: approve the suggested fixes, provide custom modification instructions, skip specific fixes, or stop the loop early. When false (default), the loop runs fully autonomously.
• COMPACT = false — When true, (1) read EXPERIMENT_LOG.md and findings.md instead of parsing full logs on session recovery, (2) append key findings to findings.md after each round.
• REVIEWER_DIFFICULTY = medium — Controls adversarial depth: medium uses normal Codex xhigh review through spawn_agent / send_input; hard adds Reviewer Memory and Debate Protocol; nightmare adds direct repository-reading adversarial verification by an independent reviewer.
• RENDER_HTML = true — When true (default), auto-render review-stage/AUTO_REVIEW.md to HTML on loop termination via /render-html. Uses --no-review (the loop itself IS the cross-model review; the HTML render is a structural conversion). Set false to skip, or pass — render html: false.

💡 Override: /auto-review-loop "topic" — compact: true, human checkpoint: true, difficulty: hard

Claude-Aligned Reviewer Memory and Debate

For difficulty: hard and difficulty: nightmare, maintain review-stage/REVIEWER_MEMORY.md.

• Before each reviewer call, prepend the full REVIEWER_MEMORY.md contents under ## Your Reviewer Memory (persistent across rounds).
• Tell the reviewer to check whether prior suspicions were genuinely addressed or merely sidestepped.
• Require a Memory update section in the reviewer response.
• After Phase B, copy the Memory update into REVIEWER_MEMORY.md before writing REVIEW_STATE.json.
• In nightmare, launch an additional fresh adversarial reviewer with direct repository/file-reading instructions. It should read NARRATIVE_REPORT.md or review-stage/AUTO_REVIEW.md for the author's claims, then verify those claims against code, logs, result files, and paper drafts instead of trusting executor summaries.

Instructions

In hard and nightmare modes, the reviewer must actively look for omissions, unsupported claims, cherry-picked evidence, metric mistakes, and weaknesses the executor may have downplayed.

For difficulty: hard and nightmare, use the Debate Protocol after a critical review:

1. Codex writes a concise rebuttal with evidence, not spin.
2. Send the rebuttal to the same reviewer via send_input.
3. The reviewer rules which objections are resolved, unresolved, or newly discovered.
4. Only mark a concern resolved when the reviewer accepts the rebuttal.

State Persistence (Compact Recovery)

Long-running loops may hit the context window limit, triggering automatic compaction. To survive this, persist state to review-stage/REVIEW_STATE.json after each round:

{
  "round": 2,
  "agent_id": "019cd392-...",
  "status": "in_progress",
  "last_score": 5.0,
  "last_verdict": "not ready",
  "pending_experiments": ["screen_name_1"],
  "timestamp": "2026-03-13T21:00:00"
}

Write this file at the end of every Phase E (after documenting the round). Overwrite each time — only the latest state matters.

On completion (positive assessment or max rounds), set "status": "completed" so future invocations don't accidentally resume a finished loop.

Workflow

Initialization

1. Check for review-stage/REVIEW_STATE.json (fall back to ./REVIEW_STATE.json if not found — legacy path):
   • If neither path exists: fresh start (normal case, identical to behavior before this feature existed)
   • If it exists AND status is "completed": fresh start (previous loop finished normally)
   • If it exists AND status is "in_progress" AND timestamp is older than 24 hours: fresh start (stale state from a killed/abandoned run — delete the file and start over)
   • If it exists AND status is "in_progress" AND timestamp is within 24 hours: resume
     • Read the state file to recover round, agent_id, last_score, pending_experiments
     • Read review-stage/AUTO_REVIEW.md to restore full context of prior rounds (fall back to ./AUTO_REVIEW.md)
     • If pending_experiments is non-empty, check if they have completed (e.g., check screen sessions)
     • Resume from the next round (round = saved round + 1)
     • Log: "Recovered from context compaction. Resuming at Round N."
2. Read project narrative documents, memory files, and any prior review documents. When COMPACT = true and compact files exist, prefer findings.md + EXPERIMENT_LOG.md over full raw logs.
3. Read recent experiment results (check output directories, logs)
4. Identify current weaknesses and open TODOs from prior reviews
5. Initialize round counter = 1 (unless recovered from state file)
6. Create/update review-stage/AUTO_REVIEW.md with header and timestamp

Loop (repeat up to MAX_ROUNDS)

Phase A: Review

Route by REVIEWER_DIFFICULTY:

Medium (default) — Codex Review

Send comprehensive context to the external reviewer:

spawn_agent:
  reasoning_effort: xhigh
  message: |
    [Round N/MAX_ROUNDS of autonomous review loop]

    [Full research context: claims, methods, results, known weaknesses]
    [Changes since last round, if any]

    Please act as a senior ML reviewer (NeurIPS/ICML level).

    1. Score this work 1-10 for a top venue
    2. List remaining critical weaknesses (ranked by severity)
    3. For each weakness, specify the MINIMUM fix (experiment, analysis, or reframing)
    4. State clearly: is this READY for submission? Yes/No/Almost

    Be brutally honest. If the work is ready, say so clearly.

If this is round 2+, use send_input with the saved agent id to maintain continuity.

Hard — Codex Review + Reviewer Memory

Use the same spawn_agent / send_input route as medium, but prepend the full review-stage/REVIEWER_MEMORY.md contents under ## Your Reviewer Memory (persistent across rounds) and require a Memory update section in the reviewer response.

Nightmare — Independent Repository Review

Use everything in hard mode, then ask an additional fresh adversarial reviewer to verify claims against repository files, logs, result files, and paper drafts instead of trusting executor summaries. Preserve the fresh review as a separate raw response and trace.

Phase B: Parse Assessment

CRITICAL: Save the FULL raw response from the external reviewer verbatim (store in a variable for Phase E). Do NOT discard or summarize — the raw text is the primary record.

Then extract structured fields:

• Score (numeric 1-10)
• Verdict ("ready" / "almost" / "not ready")
• Action items (ranked list of fixes)

STOP CONDITION: If score >= 6 AND verdict ∈ {"ready", "