# Rebuttal Desk

> Answer peer review: paste the editor's decision letter and the reviewer reports, get every request and comment numbered in the browser for free, then a revision plan, reviewer-separated point-by-point responses, or an audit of your own draft reply.

Rebuttal Desk is a web app at https://rebuttal-desk.skillsafe.ai/ for authors revising a scientific manuscript. It is derived from the agent skill @yuan1z0825/nature-response (yuan1z0825/nature-skills, Apache-2.0). It is not affiliated with any journal or publisher.

## What runs free in the browser

- Decision label read only from an explicit phrase in the editor's part of the letter (Major Revision, Minor Revision, revise and resubmit, transfer, rejection). A reviewer's own recommendation never sets it; with no label the app asks.
- Manuscript id, title, deadline, required files (point-by-point response, marked-up and clean manuscript, reporting summary, data availability statement) and visibility notes (transparent peer review, responses shared with all reviewers).
- Reviewer boundaries ("Reviewer #1", "Referee 2", "Referee B") and comment numbering: editor requests E.1, E.2 and reviewer comments R1.1, R2.3, from numbered, lettered, bulleted, "Comment 3" or "Line 45:" lists, with Major and Minor sections. Summary and closing paragraphs with no request are listed but not numbered.
- Confidential comments to the editor are set aside and never sent as comments.
- Hints per comment: category (editorial, evidence, methodological, statistical, data and code, citation, scope, ethics), ethics and statistics flags, concerns two reviewers repeat.
- Author notes keyed by id ("R1.2: second shRNA done, Supplementary Fig. S5") with a status hint: reported done with or without a location, to do, disagreement.
- Draft reply lint (audit lane): comments with no reply, thanks-only or "revised accordingly" answers, changes claimed without a location, mentions of another reviewer or pointers to another response, rebukes ("already stated", "the reviewer overlooked"), time or funding excuses, placeholders, locations your inputs never mention, revised passages not verbatim in the manuscript text you pasted, and confidential wording repeated.
- LaTeX response files: a port of the skill's check_package_consistency.py (quotes in \RevisedExcerpt must be verbatim in the manuscript; \ReviewerComment and \AuthorResponse counts must match).

## Paid lanes (model gpt-terra, sign-in required)

- plan: internal master tracker, one row per id with category, severity, action (ACCEPT_TEXT, ACCEPT_ANALYSIS, CLARIFY_EXISTING, SOFTEN_CLAIM, PARTIAL, DISAGREE, OUT_OF_SCOPE, AUTHOR_INPUT_NEEDED, BLOCKING ...), work status (VERIFIED_DONE, REPORTED_DONE_UNVERIFIED, TODO_*, NOT_FEASIBLE, PROPOSED_DISAGREEMENT), required input, expected output, location and whether it blocks finalization; strategy, conflicts and a change checklist.
- draft: a response per id, assembled by the app into one standalone file per reviewer (and one for the editor) with the comment copied from the letter, as Markdown or LaTeX; optional cover letter.
- audit: coverage per id, findings with verbatim evidence from your draft, up to three rewrites.

Every reply carries a readiness: ready_to_submit, draft_with_placeholders, needs_author_input or blocked.

## What the app checks in every reply

One row per id; the decision copied, not inferred; readiness no looser than the rows; VERIFIED_DONE only with evidence copied from the pasted manuscript; "done" only where a note reports it; DISAGREE only where a note approves it; no figure, table, line, reference, identifier or decimal the inputs never mention; no reviewer-facing text that names another reviewer or an internal id, rebukes the reviewer, or gives time or money as a reason.

## Run input

task (plan, draft or audit), title, decision, letter (line-numbered), notes, manuscript, draft (audit), plan, scope and cover_letter (draft), facts (a JSON string from the browser), question, retry_note. See https://rebuttal-desk.skillsafe.ai/api.html for the API.

## Links

- App: https://rebuttal-desk.skillsafe.ai/
- API tutorial: https://rebuttal-desk.skillsafe.ai/api.html
- Source skill: https://skillsafe.ai/skill/@yuan1z0825/nature-response
- Source repository: https://github.com/yuan1z0825/nature-skills (Apache-2.0)