Research Review

Override for Codex users who want Claude Code, not a second Codex agent, to act as the reviewer. Install this package after skills/skills-codex/*.

This reviewer is a different model family from the Codex executor. Every overlay trace/audit records:

yamlCopy code
review_independence: cross-family
acceptance_status: accepted

Research Review via claude-review MCP (high-rigor review)

Claude overlay assurance: this route is a different model family from the Codex executor and records review_independence: cross-family plus acceptance_status: accepted.

Get a multi-round critical review of research work from an external LLM with maximum reasoning depth.

Constants

• REVIEWER_MODEL = claude-review — Claude reviewer invoked through the local claude-review MCP bridge. Set CLAUDE_REVIEW_MODEL if you need a specific Claude model override.
• REVIEWER_BACKEND = claude-review — reviews route through the claude-review MCP (Claude family; cross-family for a Codex executor).

Context: $ARGUMENTS

Prerequisites

• Install the base Codex-native skills first: copy skills/skills-codex/* into ~/.codex/skills/.
• Then install this overlay package: copy skills/skills-codex-claude-review/* into ~/.codex/skills/ and allow it to overwrite the same skill names.
• Register the local reviewer bridge:

  bashCopy code
  codex mcp add claude-review -- python3 ~/.codex/mcp-servers/claude-review/server.py
  

• This gives Codex access to mcp__claude-review__review_start, mcp__claude-review__review_reply_start, and mcp__claude-review__review_status.

Workflow

Step 1: Gather Research Context

Before calling the external reviewer, compile a comprehensive briefing:

1. Read project narrative documents (e.g., STORY.md, README.md, paper drafts)
2. Read any memory/notes files for key findings and experiment history
3. Identify: core claims, methodology, key results, known weaknesses

Step 2: Initial Review (Round 1)

Send a detailed prompt with ultra reasoning:

Copy code
mcp__claude-review__review_start:
  prompt: |
    [Full research context + specific questions]
    Please act as a senior ML reviewer (NeurIPS/ICML level). Start from the
    assumption that the work is broken somewhere — your job is to find where.
    Be adversarial. Trust nothing the author tells you — verify everything
    yourself. Identify:
    1. Logical gaps or unjustified claims
    2. Missing experiments that would strengthen the story
    3. Narrative weaknesses
    4. Whether the contribution is sufficient for a top venue

    === SCOPE LIMITS (these bound what you PROPOSE, never what you look for) ===
    Report anything that is actually wrong here — including a rare-looking case, if
    this repo actually produces it. Then keep the fix in scope:
    1. This is a RESEARCH-WORKFLOW tool, not a security paper. Verification is
       welcome; over-defense is not. Assume a cooperating operator on their own
       machine — a malicious local user is NOT in the threat model.
    2. Do NOT propose SHA / hash / content-fingerprint / digest-binding schemes.
       Reporting a real defect in hashing code that already exists is fine.
    3. NO speculative machinery: do not add feature flags, migration frameworks,
       compat layers, wrappers, pins, or similar mechanisms unless evidence shows
       a current repo defect they fix or an explicit existing invariant they must
       preserve. "Load-bearing", "compatibility", and "not scaffolding" are labels,
       not evidence. Point to the failing path/artifact or invariant, and check the
       proposal's factual premises, such as whether a named package version exists.
    4. NO corner-case obsession: exotic encodings, symlink races, RTL text and
       millisecond races are out of scope unless you can show the case arises here.
    5. Where a rubric or checklist is genuinely needed, do not over-mechanize
       judgement. A clear sentence a human reads beats a scored table nobody
       maintains.
    Exception: code that runs remote commands, starts a network service, or installs
    an MCP server runs on the user's machine with their credentials — trust-boundary
    findings there are in scope and the default is strict.
    Say plainly when something is correct. Do not manufacture findings.
    Be brutally honest. If, after genuinely trying to break it, the work
    holds up and is ready, say so clearly.

After this start call, immediately save the returned jobId and poll mcp__claude-review__review_status with a bounded waitSeconds until done=true. Treat the completed status payload's response as the reviewer output, and save the completed threadId for any follow-up round.

Step 3: Iterative Dialogue (Rounds 2-N)

Use mcp__claude-review__review_reply_start with the saved completed threadId, then poll mcp__claude-review__review_status with the returned jobId until done=true to continue the conversation:

Copy code
mcp__claude-review__review_reply_start:
  # the bridge grants the reviewer no tools by default; this prompt passes
  # artifact paths, so it has to ask for read-only access explicitly
  tools: "Read,Grep,Glob"
  threadId: [saved reviewer id from Step 2]
  prompt: |
    Please continue the review using the revised materials below.

    Revised files:
    - /absolute/path/to/file1
    - /absolute/path/to/file2

    Focus on unresolved weaknesses and whether the revision actually fixed them.

After this start call, immediately save the returned jobId and poll mcp__claude-review__review_status with a bounded waitSeconds until done=true. Treat the completed status payload's response as the reviewer output, and save the completed threadId for any follow-up round.

For each round:

1. Respond to criticisms with evidence/counterarguments
2. Ask targeted follow-ups on the most actionable points
3. Request specific deliverables: experiment designs, paper outlines, claims matrices

Key follow-up patterns:

• "If we reframe X as Y, does that change your assessment?"
• "What's the minimum experiment to satisfy concern Z?"
• "Please design the minimal additional experiment package (highest acceptance lift per GPU week)"
• "Please write a mock NeurIPS/ICML review with scores"
• "Give me a results-to-claims matrix for possible experimental outcomes"

Step 4: Convergence

Stop iterating when:

• Both sides agree on the core claims and their evidence requirements
• A concrete experiment plan is established
• The narrative structure is settled

Step 5: Document Everything

Save the full interaction and conclusions to a review document in the project root:

• Round-by-round summary of criticisms and responses
• Final consensus on claims, narrative, and experiments
• Claims matrix (what claims are allowed under each possible outcome)
• Prioritized TODO list with estimated compute costs
• Paper outline if discussed

Update project memory/notes with key review conclusions.

If — composed: <canonical-report-path> is explicitly present, fold consensus, claims matrix, TODOs, and trace links into that report instead of writing a standalone review document. Without the directive, write the standalone review as documented; never infer composed mode from an existing file. — standalone always wins. See output-composition.md.

Step 6: Review Tracing

Save a trace for every mcp__claude-review__review_start, mcp__claude-review__review_reply_start, or oracle-pro review call following ../shared-references/review-tracing.md. Record the reviewer route, saved threadId, prompt summary, raw response path, decisions, and action items. This preserves the Claude mainline Review Tracing semantics while using Codex-native reviewer calls.

Key Rules

• Always ask the Claude reviewer for strict, high-rigor feedback in every review round.
• Send comprehensive context in Round 1 — the external model cannot read your files
• Be honest about weaknesses — hiding them leads to worse feedback
• Push back on criticisms you disagree with, but accept valid ones
• Focus on ACTIONABLE feedback — "what experiment would fix this?"
• Document the completed threadId for potential future resumption
• The review document should be self-contained (readable without the conversation)

Prompt Templates

For initial review:

"I'm going to present a complete ML research project for your critical review. Please act as a senior ML reviewer (NeurIPS/ICML level)..."

For experiment design:

"Please design the minimal additional experiment package that gives the highest acceptance lift per GPU week. Our compute: [describe]. Be very specific about configurations."

For paper structure:

"Please turn this into a concrete paper outline with section-by-section claims and figure plan."

For claims matrix:

"Please give me a results-to-claims matrix: what claim is allowed under each possible outcome of experiments X and Y?"

For mock review:

"Please write a mock NeurIPS review with: Summary, Strengths, Weaknesses, Questions for Authors, Score, Confidence, and What Would Move Toward Accept."