Referee reports where every criticism is located, actionable, and literature-backed.
Write, structure, and audit peer-review reports for biomedical journal manuscripts: a 26-class mechanical screener, four reading passes, verdict calibration against real review-outcome base rates, house-style drafting (summary → major → minor → references), citation verification at export, and second-round revision audits. Home turf: bone/skeletal biology, matrix biology, orthodontics/craniofacial, orthopaedics, and omics.
$5.00 one-off
Preview — package delivery is not available yet
python3; pdftotext (poppler) or pypdf for PDF manuscripts; python-docx for .docx export
License and attribution corrected on 25 September 2026. Scientific demonstrations and native-agent checks retain their original dates and scope; no new independent human review is claimed.
Manuscripts under review are confidential. Most large publishers forbid or require disclosure of generative-AI use in refereeing. This skill is policy-aware by design: it opens every review by establishing the journal's AI policy and offers a reviewer-drafted mode in which only your own rough comments are processed — never silently assume permission.
The skill's coverage was calibrated against 60 published open peer reviews (45 PLOS ONE + 15 F1000/Wellcome, bone/orthopaedics/matrix biology, via Europe PMC): 1,035 reviewer comments coded into a 20-class taxonomy, then checked class-by-class against the skill's passes and checklists.
Result: every taxonomy class is prompted by the skill; the largest residual gap found (design-parameter justification) was closed with a dedicated pass before this edition shipped. Real reviewers' verdicts in the corpus ran minor-or-better 73% of the time — the skill's verdict calibration is tuned to that base rate rather than to over-issuing major revisions.
On 2026-08-12 the skill was run blinded on the original submitted version of a open-review wet-lab article (not in the calibration corpus), then scored against the published referee reports and the actual v1→v2 revisions, with an adversarial audit of every headline match.
The retained PeerJ run found the unsupported 'Method 2 is best' ranking, matched the editor's Major Revision verdict, demanded 5 of 9 review-driven revisions, and added 3 audited findings. Six row-level overlaps became four substantive matches after audit. A second, contaminated arm was withdrawn and is excluded. Its measured weakness was breadth on bench-protocol details and figure conventions.
The complete, unedited referee report the skill produced blinded on the wet-lab article's v1. Download it, then open the real reviewers' reports (published under CC BY on the article's page) and the authors' v2 amendments, and judge the comparison yourself.
Generated as a capability demonstration, not submitted to any journal; its concerns are phrased as questions, and the article's own open review raised — and the authors answered — several of the same points. The wet-lab run's full review is downloadable too.
29 files · 271 KiB unpacked · standard Agent Skills layout
Package delivery is not available yet. These instructions describe installation once you have the ZIP. The compatibility table records the actual test date, environment and scope for each agent.
Installation overview. Follow the steps for your agent and check the compatibility table for the tested scope.
Checkout and package delivery are not available yet. These are installation instructions for a future downloaded ZIP.
Preview — package delivery is not available yet
The steps below describe installation once you have the ZIP. Checkout and downloads are not enabled.
Personal scope shown; use <project>/.claude/skills/ for project scope. Claude.ai (web) accepts the zip directly under Settings → Skills, but cannot run the bundled Python tooling.
mkdir -p ~/.claude/skills/academic-peer-review
unzip academic-peer-review-1.0.2.zip -d ~/.claude/skills/academic-peer-reviewA new session discovers the skill automatically from its SKILL.md description.
claude -p "Do you have the academic-peer-review skill available? State its name and what it does."| Agent | Status | Method | Notes |
|---|---|---|---|
| Claude Code 2.1.209 | Verified2026-08-12 | Native discovery from ~/.claude/skills / <project>/.claude/skills | The skill was developed in and is used daily in Claude Code; this catalog's curation ran in it. |
| Codex CLI 0.146.0 | Verified2026-08-12 | Native discovery from ~/.codex/skills | Smoke test passed: named the skill, quoted its mandated process, and named its bundled script — all read from the installed package. |
| Antigravity (agy) 1.1.8 | Partial2026-08-12 | Explicit pointer: tell the agent to read SKILL.md | No native skills-directory injection observed in headless mode; interactive sessions load it by pointer. In the AROS stack, ~/.gemini/skills is surfaced by the AROS layer. |
| claude.ai (web) — | Untested | Zip upload under Settings → Skills | The package matches the upload format, but the web sandbox cannot run the bundled Python tooling. |
| OpenClaw — | Untested | Standard Agent Skills layout | Not available on our test bench. If you verify it, tell us and we'll record it here. |
Organization-use commercial license for Nexitia material, with third-party publication rights preserved. Outputs may be published and used commercially. Earlier grants are preserved.
The displayed price does not change the license. Terms apply to the identified edition; earlier grants and third-party rights are preserved.
Encodes the curator's own refereeing practice, distilled from 26 review documents (2017–2026) for bone, matrix-biology, craniofacial and orthopaedics journals, with worked examples fictionalized so no reviewed manuscript is identifiable.
SHA-256 of academic-peer-review-1.0.2.zip — verify your download against this hash:
f86a6ee9652bb339c3daf3f55dc520680105712d2ded9f392a8e40d533e37dc2