/review

name: review
description: All quality reviews — routes to appropriate critics based on target file type and flags. Replaces /paper-excellence, /proofread, /econometrics-check, /review-r, /review-paper.
argument-hint: "[file path or --flag] Options: --peer [journal], --stress [journal], --methods, --proofread, --code, --replicate [lang], --all"
allowed-tools: Read,Grep,Glob,Write,Bash,Task

Review

Unified review command that routes to the appropriate critic agents based on the target and flags.

**Input:** `$ARGUMENTS` — file path and/or flags.

---

Routing Logic

Auto-detect by file type

• `.tex` paper file → **Comprehensive review** (writer-critic + strategist-critic + Verifier)
• `.R`, `.py`, `.do`, `.jl` file → **Code review** (coder-critic standalone, categories 4-12)
• `.tex` talk file (in talks/) → **Talk review** (storyteller-critic)

Explicit flags (override auto-detect)

• `--peer [journal]` → **Full peer review** (editor desk review → referee dispatch → editorial decision)
• `--peer --r2 [journal]` → **R&R second round** (same referees, same dispositions, memory of prior review)
• `--stress [journal]` → **Hostile stress test** (same flow, adversarial referee dispositions)
• `--methods` → **Causal audit** (strategist-critic standalone, 4-phase review)
• `--proofread` → **Manuscript polish** (writer-critic standalone, 6 categories)
• `--code [file]` → **Code review** (coder-critic standalone, categories 4-12)
• `--replicate [language]` → **Cross-language replication** (Coder re-implements in target language + coder-critic + comparison)
• `--all` or no file → **Paper excellence** (all critics in parallel + weighted score)

---

Mode Details

Comprehensive Review (default for .tex paper)

Dispatch in parallel: 1. **strategist-critic** — causal design audit (4 phases) 2. **writer-critic** — manuscript polish (6 categories) 3. **Verifier** — compilation check Compute weighted aggregate score.

Full Peer Review (`--peer [journal]`)

Simulates a realistic journal submission. Three phases, orchestrated sequentially.

Phase 1: Editor Desk Review

Dispatch the **editor** agent with the paper and target journal.

The editor: 1. Reads the paper (abstract, intro, contribution, identification, results) 2. Searches the literature via WebSearch to verify novelty claims 3. Decides: **DESK REJECT** or **SEND TO REFEREES** 4. If desk reject → report with reasons + suggested alternative journals. Done. 5. If send to referees → editor selects referee dispositions and pet peeves from the journal's **Referee pool** (see .claude/references/journal-profiles.md)

Phase 2: Referee Reports

The editor's referee assignment specifies for each referee:

• **Disposition** (one of: STRUCTURAL, CREDIBILITY, MEASUREMENT, POLICY, THEORY, SKEPTIC)
• **Critical pet peeve** (one from the critical pool)
• **Constructive pet peeve** (one from the constructive pool)

Dispatch **domain-referee** and **methods-referee** in parallel, each receiving: 1. The paper manuscript 2. The target journal name (for .claude/references/journal-profiles.md calibration) 3. Their assigned disposition and pet peeves, injected into the prompt:

DISPOSITION: [disposition name]
You approach this paper with the following intellectual prior: [disposition description]
This shapes your emphasis, not your scoring rubric — the 5 dimensions remain the same.

PET PEEVES:
- Critical: [critical pet peeve]
- Constructive: [constructive pet peeve]
Give extra weight to these in your review. The critical peeve is something you particularly
care about and will scrutinize. The constructive peeve is something you appreciate and will
reward when present.

Both reviews are independent and blind — neither referee sees the other's report.

Every major comment MUST include a **"What would change my mind"** statement — not just "this is wrong" but the specific evidence, test, or analysis that would resolve the concern.

Phase 3: Editorial Decision

Dispatch the **editor** agent again with both referee reports.

The editor: 1. Classifies each concern as FATAL / ADDRESSABLE / TASTE 2. When referees disagree, takes a side and explains why 3. Produces a decision letter: Accept / Minor Revisions / Major Revisions / Reject 4. Lists MUST address, SHOULD address, and MAY push back items

Save Reports

Save all outputs to `quality_reports/reviews/`:

• `YYYY-MM-DD_desk_review.md` (Phase 1)
• `YYYY-MM-DD_referee_domain.md` (Phase 2)
• `YYYY-MM-DD_referee_methods.md` (Phase 2)
• `YYYY-MM-DD_editorial_decision.md` (Phase 3)

Log the referee assignments (dispositions + pet peeves) in the editorial decision so the user can re-run with different combinations.

R&R Second Round (`--peer --r2 [journal]`)

Continues the review cycle after the author has revised the paper.

1. **Load prior review state** — read previous referee reports and editorial decision from `quality_reports/reviews/` 2. **Skip desk review** — the paper was already accepted for review 3. **Same referees** — reload the same dispositions and pet peeves from round 1 4. **Referee R&R mode** — each referee receives their previous report alongside the revised manuscript:

You previously reviewed this paper. Your prior report is attached.
Check whether each concern you raised has been adequately addressed.
New concerns may arise from the revisions. Score the revision, not
the original — improvement matters.

They check whether each concern was: Resolved / Partially resolved / Not addressed. They may flag new concerns from the revisions.

5. **Editor R&R decision** — Round 2 allows Accept/Minor/Major/Reject. Round 3 allows Accept/Minor/Reject only. Max 3 rounds total — editor's patience runs out, just like real life. 6. **Save reports** with `_r2` or `_r3` suffix to `quality_reports/reviews/`

Hostile Stress Test (`--stress [journal]`)

Same three-phase flow as `--peer`, with these changes:

1. **Editor assigns adversarial dispositions** — both refere