From reviewer-author-loop
Iterative manuscript improvement workflow where a reviewer agent critiques, an author agent revises, a verifier checks the revision, and the loop repeats until acceptance or a human-pause condition is reached. Use for virtual peer review, reviewer-author revision loops, rebuttal preparation, resubmission polishing, journal paper improvement, response-to-reviewers drafting, and deciding when additional data, experiments, theory, modeling, or author judgment is required before further revision.
How this skill is triggered — by the user, by Claude, or both
Slash command
/reviewer-author-loop:reviewer-author-loopThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Use this skill to run a closed-loop manuscript improvement process:
Use this skill to run a closed-loop manuscript improvement process:
This skill is distinct from one-pass peer review. Its value is the revision loop, traceable comment resolution, and explicit stop/pause decision.
Default behavior is a full automated loop, not a one-pass review. After the Reviewer Agent produces comments, immediately continue to the Author Agent unless one of these is true:
Do not end the turn after "Reviewer Round 1" if the user asked for review-revise-review, iterative improvement, or improvement until acceptance. If the user asks to "start with Reviewer Round 1," provide a concise reviewer summary, then continue automatically into author revision, verification, and re-review.
For long manuscripts, run one complete loop per turn when feasible:
Reviewer Round N -> Author Revision N -> Verifier Check N -> Re-Reviewer Decision N
Only stop at the end of a loop when the stop/pause rules below are satisfied. Otherwise continue to the next loop or explain which practical limit prevents continuing.
If private manuscripts, reviewer reports, review archives, unpublished data, grant proposals, or confidential comments are provided, use them only for the current task. Do not copy private text into reusable skill files, public outputs, examples, or repositories. When extracting lessons from private materials, convert them into abstract process rules with no titles, author names, manuscript identifiers, wording, or identifiable facts.
Use this skill when the user asks to:
If the manuscript is in a specific technical field and a domain skill is available, use this skill as the process scaffold and the domain skill as the judgment layer.
Before starting, identify what is available:
Proceed with reasonable assumptions when the missing information is not blocking. Pause only when the missing information changes the revision path.
Read the current manuscript and produce a decision-oriented review:
Use references/reviewer-rubric.md for detailed review dimensions.
Keep reviewer output concise enough to guide revision. The reviewer report is an intermediate artifact unless the user requested review-only output.
Convert reviewer comments into revisions:
Use references/author-revision-protocol.md for revision strategy and response-log structure.
If source files are editable, make the revisions directly. If direct editing is not possible, produce a revised section, patch plan, or response log as the author artifact.
Check the revision before returning to review:
If editing files, do not claim completion until verification has been run or the inability to run it is reported.
Re-review the revised manuscript against the previous acceptance criteria:
If more revisions are possible with available information, loop back to the Author Agent. If not, pause.
Stop when the reviewer recommendation is:
Pause and ask the user when progress requires:
Use references/pause-criteria.md to classify pause conditions and phrase the user request.
During an automated loop, do not dump full reviewer reports at every stage unless asked. Keep outputs concise and actionable:
For intermediate loops, include a compact response log:
ID | Reviewer concern | Author action | Location/artifact | Verification status
When the loop stops, provide:
In the manuscript workflow that motivated this skill, the reviewer-author loop acted as the process scaffold while other skills supplied specialist judgment or file operations.
When acting as the Reviewer Agent, use and cite companion skills as applicable:
academic-paper-reviewer: independent peer-review stance, editorial recommendation, major/minor comments, re-review after revision.mechanical-engineering-research: domain-specific technical review for thermal-fluid physics, boiling/heat-transfer models, experimental design, uncertainty, equations, figures, and interpretation.deep-research: targeted literature checks when a review comment depends on current or unfamiliar literature.When acting as the Author Agent, use and cite companion skills as applicable:
academic-paper: manuscript restructuring, abstract/introduction/discussion revision, response-to-reviewer style writing, and paper-level coherence.mechanical-engineering-research: technical rewriting, mechanistic framing, model explanation, analysis design, and engineering interpretation.documents:documents: DOCX editing, rendering, visual QA, comments, and tracked document artifact work.spreadsheets:Spreadsheets: spreadsheet inspection, data extraction, recalculation, and analysis tables.zotero:Zotero: citation lookup, BibTeX export, reference cleanup, and citation-key insertion when Zotero or BibTeX is used.presentations:Presentations: figure or slide-deck integration when manuscript figures are being adapted from presentation material.For one-pass critique only, a peer-review skill is sufficient. Use this skill when the user wants iterative review, revision, verification, and re-review. In outputs, name the companion skills used for each loop stage so the workflow remains auditable.
flowchart TD
A["Manuscript + target journal"] --> B["Reviewer critiques"]
B --> C["Decision + required revisions"]
C --> D{"Can revise with available information?"}
D -- "Yes" --> E["Author revises manuscript"]
E --> F["Verifier checks build, citations, figures, claims"]
F --> G["Reviewer re-reviews"]
G --> H{"Acceptable?"}
H -- "Yes" --> I["Stop: ready or near-ready"]
H -- "No, revision possible" --> C
D -- "No" --> J["Pause for human input"]
J --> K["Human provides data, decisions, or theory"]
K --> E
Guides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.
npx claudepluginhub hanhuark/reviewer-author-loop-skill