Scientific Peer Review

Peer Review is a skill in K-Dense's Claude Scientific Writer (MIT) that supports an accountable human reviewer. Version 2.1 treats every unpublished submission as confidential: it will not analyze a manuscript until authorization, conflicts, and venue AI policy are recorded, and it never announces an editorial decision.

Key Features

• Intake gate: you fill assets/review_intake_template.json and run validate_review_intake.py. Work proceeds only at status READY_FOR_LOCAL_REVIEW. Undocumented authorization, unresolved conflicts, unchecked venue policy, external service use, or missing deletion plans block it.
• 11-step workflow: scope, neutral orientation, reporting-guideline selection, claim-to-evidence mapping, methods and statistics, reproducibility, ethics and integrity, figures and citations, actionable comments, separate channels for authors and editor, then lint and finalize.
• Seven local CLIs: intake validation, guideline selection with coverage audit, a claim-evidence matrix check, a statistics and reproducibility checklist, a Pandoc citation-key audit, a review scaffold generator, and a linter. They are deterministic, make no network or model calls, and report IDs and line numbers rather than echoing manuscript text.
• Comment format: each major or minor point needs location, observation, evidence or criterion, why it matters, and the requested action.

Use Cases

• Reviewing a journal manuscript or preprint you were invited to assess.
• Evaluating a grant proposal or protocol for design and reporting gaps.
• Planning a response to reviewers by stress-testing your own draft first.

Pricing and Access

Free under MIT. It needs only the Python 3.11 standard library, and runs offline by design, so there is no API cost beyond your model session.

Getting Started

1. Install the plugin in Claude Code: /plugin marketplace add https://github.com/K-Dense-AI/claude-scientific-writer, then /plugin install claude-scientific-writer.
2. Complete the intake file and run python3 scripts/validate_review_intake.py completed-intake.json.
3. Generate a private draft with python3 scripts/generate_review_scaffold.py completed-intake.json -o private-review.md, write it, then python3 scripts/lint_review.py private-review.md.

Limitation: automated checks are not peer review and do not measure merit. The skill forbids sending unpublished text to external services without publisher or author authorization, so check the venue policy before using any hosted model.

Frequently Asked Questions

Will it recommend accept or reject?

No. Decisions belong to editors or panels; the linter even flags decision phrases.

Can it reproduce the authors' analysis?

Only if you run authorized inputs yourself with documented commands. The skill forbids claiming reproduction otherwise.

Alternatives

• Scholar Evaluation Skill: structured quality assessment of scholarly work.
• Scientific Critical Thinking Skill: focuses on reasoning, bias, and evidence strength.
• Scientific Writing Skill: use on your own manuscript before others review it.

Conclusion

Peer Review is a disciplined scaffold for careful reviewers who need confidentiality and accountability built in. It will slow you down on purpose. Browse more research skills in the skills hub.