rebuttal-guidance — Claude Code skill: plan your reviewer response
A Claude skill that splits a peer review into numbered concerns and hands you a response plan for each. You still write the final prose.
npx skills add HKUSTDial/Supervisor-Skills --skill rebuttal-guidance -a claude-codeWhat rebuttal-guidance does
Paste a review and the skill lists every numbered weakness and question, then sorts each one into a category: misunderstanding, missing evidence, novelty, baseline fairness, scope, clarity of writing, theory, efficiency. The categories decide what to answer first.
What drives the advice is a conditional-persuasion model. The skill reads the reviewer's tone from the review text, assigns a mindset cluster, and states how confident that call is. Each cluster comes with historical up-rate figures and a comparison of winning and losing responses. It also points to ten strategy dimensions and three response templates, A, B and C.
No paste-ready paragraphs. For every concern you get a skeleton — acknowledge, answer, show evidence, note the revision — plus a pointer to the section, table or appendix that backs it up and a yes/no on whether fresh experiments are needed. Made-up numbers are off the table; anything unchecked is marked [TO VERIFY]. A review text is the starting point, so without one the skill asks for it. Judging a research idea or reviewing a paper before submission is outside its job: the same set ships separate skills for that.
What it helps with
- turn a messy review into a numbered list of concerns
- tag each concern by type before drafting
- pick tactics that fit a hostile or an open reviewer
- locate the table or appendix that answers a question
- run a pre-submit integrity check on the response
Example prompts for the agent
- Break this review into concerns and give me a plan for each one.
- How should I answer R1.W2 when the reviewer calls my baseline unfair?
- The reviewer ignored my response — recalibrate the strategy for their mindset.
How to install rebuttal-guidance
1Skills CLI (Vercel)
npx skills add HKUSTDial/Supervisor-Skills --skill rebuttal-guidance -a claude-code2Manually: put the files in place
~/.claude/skills/rebuttal-guidance/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/rebuttal-guidance/
1Skills CLI (Vercel)
npx skills add HKUSTDial/Supervisor-Skills --skill rebuttal-guidance -a codex2Manually: put the files in place
.agents/skills/rebuttal-guidance/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add HKUSTDial/Supervisor-Skills --skill rebuttal-guidance -a cursor2Manually: put the files in place
.cursor/skills/rebuttal-guidance/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install HKUSTDial/Supervisor-Skills rebuttal-guidance2Manually: put the files in place
.github/skills/rebuttal-guidance/SKILL.md1Skills CLI (Vercel)
npx skills add HKUSTDial/Supervisor-Skills --skill rebuttal-guidance -a gemini-cli2Platform CLI
gemini skills install https://github.com/HKUSTDial/Supervisor-Skills --path skills/rebuttal-guidanceManually: put the files in place
.devin/skills/rebuttal-guidance/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add HKUSTDial/Supervisor-Skills --skill rebuttal-guidance -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about rebuttal-guidance
How do I install rebuttal-guidance in Claude Code?
Run npx skills add HKUSTDial/Supervisor-Skills --skill rebuttal-guidance -a claude-code. The skill lands at ~/.claude/skills/rebuttal-guidance/SKILL.md.
What does rebuttal-guidance do?
It inventories the review, classifies each concern and returns a per-concern response structure with tone advice. It does not produce finished rebuttal text.
Can I paste its output into OpenReview as is?
No. The skill supplies plans and templates; the author writes the text. It also warns against dropping a raw model answer into the thread.