Peer Review

What it does

Turns manuscript review into a structured, evidence-anchored process rather than impressionistic commentary.

• Safety boundary first: requires confirming reviewer authorization, checking the venue's confidentiality/AI/co-review policies, and recording conflicts and competence limits before any unpublished text is read. Explicitly forbids uploading confidential content to external models, plagiarism checkers, or citation services.
• Intake gate: validate_review_intake.py blocks progress until the declaration reaches READY_FOR_LOCAL_REVIEW.
• 10-step workflow: scope and available evidence → neutral orientation map → reporting-guideline selection (CONSORT, PRISMA, etc.) → claim-to-evidence mapping → methods and statistics → reproducibility and transparency → ethics and integrity → figures, tables, citations → actionable comments → author/editor channel separation → lint and finalize.
• Seven deterministic local CLIs: guideline selector with non-scoring coverage audit, claim–evidence matrix validator, statistics/reproducibility auditor, Pandoc citation auditor, private review scaffold generator, and a linter for tone, channel leakage, placeholders, and actionability fields. Reports emit IDs, counts, and line numbers — never manuscript text.

Who it's for

• Researchers and faculty who regularly accept review invitations
• PIs standardizing internal preprint review or training students in review craft
• Authors self-auditing a manuscript before submission or before a response letter
• Editorial board members managing review quality and confidential-channel hygiene

Example uses

1. Handling an invited review: complete the intake template, run the validator to document authorization and conflicts, generate a CONSORT coverage table for an RCT, and draft major/minor comments in Location–Observation–Evidence–Why it matters–Requested action form.
2. Claim–evidence alignment audit: catch abstract-level causal language that outruns the reported effect size, uncertainty, or timepoint by filling the claim-evidence CSV and running validate_claim_evidence.py.
3. Pre-submission self-review: audit your own reproducibility checklist (seeds, package versions, accession IDs) and citation-key consistency to remove common desk-reject triggers.

Install the skill I found on Claude Skill Mart.
Copy the skills/peer-review folder from the GitHub repo K-Dense-AI/scientific-agent-skills into my ~/.claude/skills/peer-review/.
When it's done, tell me in one line what this skill can do.