Guide · Agent skill
How to use the Submission Rebuttal Pipeline agent skill (rebuttal)
Updated · written with AI from the project's own documentation and checked against our catalogue data
The rebuttal skill is an agent skill for Claude Code or another SKILL.md-compatible agent. It guides the agent through answering peer reviews on a paper you have already submitted, aimed mainly at ICML-style text-only rebuttals with strict character limits. After this guide you will know how to install the skill, how to ask for it in plain language, what inputs it needs, and what checks it applies before a draft is finalised.
Submission Rebuttal Pipelinewanshuiyin/Auto-claude-code-research-in-sleep · MIT Download the skill Quick start
- 1Install the skill. Unzip the downloaded ZIP into ~/.claude/skills/ for personal use or into the project's .claude/skills/ folder, or upload it under Settings → Capabilities → Skills in the Claude apps.
- 2Gather inputs. Collect the paper source, the raw reviews with reviewer IDs, the venue name and character limit, and whether this is an initial or follow-up round.
- 3Ask for the issue board first. Make a natural request and ask for the issue board and strategy plan before any drafting, or enable quick mode so the run stops after those two files.
- 4Check the issue board. Read ISSUE_BOARD.md and confirm every reviewer concern is listed with a severity, stance and status.
- 5Resolve blockers. Answer any questions the skill raises and supply any confirmed results or approved commitments it needs; it will not guess missing venue rules or results.
- 6Review the draft. Read the draft against the paper and reviews, check it stays within the character limit, and edit any claim you cannot verify.
- 7Submit yourself. Paste the final text into the venue's system. The skill does not submit anything for you.
What an agent skill is and what this one makes the agent do
An agent skill is a folder containing a SKILL.md file that tells a compatible agent how to handle one kind of task. When the request matches the skill's description, the agent follows the steps written in that file.
This skill covers post-submission peer review. It makes the agent parse each reviewer's comments, split them into individual concerns, classify each one, and plan a response strategy. It then drafts a text-only rebuttal within the character limit you give, checks that every factual claim traces to the paper, the reviews, or results you have confirmed, and checks that no reviewer concern is silently dropped. It can also manage follow-up rounds with the same reviewers.
It does not run new experiments automatically, does not edit or upload a revised PDF, and does not submit anything to OpenReview, CMT or HotCRP. You submit the final text yourself.
Install the rebuttal skill
Download the skill's ZIP from our page and unzip it. You can install it for yourself or for one project. For a personal install, unzip it into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder.
If you use the Claude apps instead of the command line, open Settings, go to Capabilities, then Skills, and upload the ZIP there.
The source package contains a SKILL.md file. The skill folder is named rebuttal, matching the skill name in that file.
- Personal: ~/.claude/skills/
- Per project: .claude/skills/ inside the project folder
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
Trigger the skill with natural requests
You do not need a special command. The skill's description lists phrases that should activate it: "rebuttal", "reply to reviewers", "ICML rebuttal", "OpenReview response", or a request to answer external reviews safely. Say what you need in ordinary words and include the paper and the reviews.
The source also shows an argument hint, [paper-path-or-review-bundle], and override settings written in the form of a command. Those are the skill's own syntax, so use them only if your agent accepts that form. Otherwise, describe the venue and limit in your message.
- "Help me write a rebuttal for ICML. The limit is 5000 characters, text only. Reviews are in reviews.md and the paper is in paper/."
- "Reply to reviewers for my OpenReview submission. I want to see the issue board before any drafting."
- "Answer these external reviews safely. Do not promise any new experiments."
What you need before starting
The skill expects four inputs. It asks you to give the paper source, the raw reviews, the venue rules, and the current stage. The paper source can be a PDF, a LaTeX directory, or a narrative summary. Reviews can be pasted text, markdown, or a PDF with reviewer IDs.
Venue rules must be explicit: the venue name, the character or word limit, and whether a text-only or revised PDF response is allowed. If the venue rules or limit are missing, the skill is designed to stop and ask before drafting.
- Paper source: PDF, LaTeX directory, or summary
- Raw reviews with reviewer IDs
- Venue name, character or word limit, and whether text-only is required
- Stage: initial rebuttal or follow-up round
Understand the outputs and safety gates
The skill writes its working files into a rebuttal/ folder. The source names REBUTTAL_STATE.md, REVIEWS_RAW.md, ISSUE_BOARD.md and STRATEGY_PLAN.md, and the rich draft is saved as REBUTTAL_DRAFT_rich.md. With QUICK_MODE on, the run stops after ISSUE_BOARD.md and STRATEGY_PLAN.md, so you can read the reviewer requests before deciding how to reply.
Three gates must pass before the skill finalises a draft. The provenance gate requires every factual statement to map to the paper, a review, or a user-confirmed result or derivation. The commitment gate requires every promise to be already done, approved for the rebuttal, or future work only. The coverage gate requires every reviewer concern to end as answered, deliberately deferred, or needing your input.
- ISSUE_BOARD.md: one entry per atomic concern with severity, stance and status
- STRATEGY_PLAN.md: shared themes, response mode per issue, character budget
- REBUTTAL_DRAFT_rich.md: the draft, rendered to HTML by default when RENDER_HTML is true
Safety notes: scripts, network and reviewer backends
The skill's allowed tools include Bash, so it can run scripts, and the source states that it uses the network. Read the scripts before you run the skill on sensitive work, and be aware that your paper and reviews may be sent to the reviewer backends it calls.
The skill can use optional reviewer backends. The listed tools are mcp__grok, mcp__antigravity and mcp__claude-review. The source notes these may need separate setup. The default internal stress-tester is a Codex MCP route, and the source describes a reviewer-routing reference file that governs how reviewers are selected. If a selected reviewer is unavailable, the skill records it as REVIEW_UNAVAILABLE rather than inventing a result.
- Runs scripts: yes
- Uses network: yes
- Optional backends may need their own setup before they work
What people use it for
- Drafting a text-only ICML-style rebuttal under a strict character limit for a paper with several reviewers.
- Building an issue board that lists every reviewer concern, so none is overlooked when replying.
- Planning which concerns to answer directly and which need new evidence or a narrow concession.
- Preparing a follow-up reply to the same reviewers across later rounds, up to the limit the skill sets per reviewer thread.
- Checking an existing draft for claims that do not trace back to the paper, the reviews, or confirmed results.
Frequently asked questions
Is the rebuttal skill free to use?
The source page does not state a price or licence terms for this skill's use beyond its listed licence, which is MIT. Check the repository licence file for the terms that apply to your copy.
What do I need to run the rebuttal skill?
You need Claude Code or another SKILL.md-compatible agent, your paper source (PDF, LaTeX directory, or summary), and the raw reviews. The skill also lists optional reviewer backends such as Codex MCP, which may need separate setup.
Will the skill submit my rebuttal to OpenReview?
No. The skill states that it does not submit to OpenReview, CMT or HotCRP, and it does not edit or upload a revised PDF. You paste the final text into the venue system yourself.
Can the skill run new experiments for me?
Not automatically. By default it pauses and shows you the evidence gaps. An optional setting can call an experiment step, but the source says that is off by default.
What if I do not know the venue's character limit?
The skill requires explicit venue rules and is designed to stop and ask rather than guess a limit before drafting.
Does the skill make sure the rebuttal is accurate?
It applies provenance, commitment and coverage checks to block ungrounded claims, but the source says you should still review every draft before submitting.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use this skill if you are answering several reviewers under a text-only, character-limited rebuttal and want full issue coverage and a structured plan. Pick something else, or write by hand, if the venue allows a revised PDF or new experiments, since the skill only drafts text, or if you need the tool to generate new results.
Download the skill