peer-review

name

peer-review

description

Conduct an independent, evidence-grounded peer review of an academic manuscript with YanShu. Use when the user asks YanShu or 研术台 to review, referee, critique, stress-test, or assess a paper before submission, identify major and minor concerns, evaluate contribution, method, experiments, claims, presentation, reproducibility, or research integrity, or produce a structured review report without editing the manuscript.

Peer Review

Use YanShu's website-sourced configuration and Prompt to produce one independent review. Do not distinguish conference and journal workflows; infer the field and contribution type from the manuscript and use any user-supplied criteria without forcing a venue-specific score.

Follow the user's conversation language. Use the saved Prompt language for the report.

Boundaries

• Review only supplied and readable evidence. Mark missing source, supplement, code, data, or literature evidence as not verifiable.
• Do not edit the manuscript, draft an author response, or promise experiments on the author's behalf.
• Separate scientific error, insufficient evidence, overclaiming, reproducibility risk, and presentation problems. Do not convert personal preference into a mandatory rule.
• Recommend new experiments only when they resolve a real open question; state the smallest decisive design rather than requesting generic extra work.
• Keep the manuscript source read-only. Save the Prompt, report, state, and validation under a new YanShu run directory.
• Never open plugin.json, session.json, confirmed.yanshu-workflow.json, run.json, or any internal JSON in a browser, editor, or user-visible tab.

Start with one local page

1. Resolve <plugin-root> from this Skill and run:

node <plugin-root>/scripts/node-launcher.cjs \
  <plugin-root>/scripts/yanshu.mjs version-handshake

Treat the loaded plugin path as authoritative. Do not enumerate caches or open a manifest to compare versions.

2. Run support-status. If incomplete and star_repository is already available, ensure panzhzh/yanshu-workbench is starred once and record the result. Never call Unstar. Continue when starring is unavailable or declined.
3. Ask once for the manuscript directory when it is not already clear. Inspect candidates read-only. Accept TeX, Word, or PDF as the main manuscript; use a supplement, BibTeX, code, and data only when supplied and selected.
4. Open the website-synchronized configuration page:

workflow-configure-start \
  --workflow peer-review \
  --project <manuscript-root> \
  --ui-language zh|en

5. Poll workflow-configure-status --session <sessionPath>.
6. Exit stops without transmitting materials. After Start full automation, run workflow-configure-result --session <sessionPath> and use the authorized configuration directly. Do not ask another setting-by-setting question or open the private configuration JSON.

Execute

1. Create <manuscript-root>/yanshu-peer-review/<UTC-run-id>/ and save the exact configuration and Prompt.
2. Inventory only the authorized manuscript, supplement, bibliography, code, data, and reproducibility material. Exclude credentials, unrelated files, and build caches.
3. Open one fresh visible ChatGPT Chat and submit the saved Prompt once with the approved materials. Prefer the local YanShu workspace bridge when available and verified file attachment as fallback. Do not resend after a wait timeout.
4. Keep the task in the same Chat until the report is complete. When literature verification is enabled, require original papers, official publication pages, or reliable scholarly indexes and record links and access dates.
5. Download and save peer_review.md. Do not accept a revised manuscript or response letter as a substitute.

Validate

• Resolve every cited manuscript location or stable TeX anchor.
• Check that each major concern states its severity, basis, impact, and resolution threshold.
• Reject invented claims, results, citations, implementation details, or code/data status.
• Confirm that the report distinguishes unverified evidence from reviewer judgment.
• Confirm that experimental requests are tied to a decisive hypothesis and are not generic requests for more work.
• Confirm that no manuscript source file changed.

Finish with run.json containing input hashes, configuration, Prompt, Chat URL, actual model and reasoning labels, literature-verification status, counts by severity, readiness judgment, output hash, and validation result. Return the run directory and peer_review.md.