triage-reviews
Skill59 starsChanged 5 months agoApache-2.0View on GitHub
Turns raw peer reviews into a prioritized triage matrix before any rebuttal is written. Use it when reviews come back from OpenReview, EasyChair, CMT, or HotCRP and the researcher says "my reviews are in", "triage these reviews", "how do I respond to Reviewer 2", "plan my rebuttal", or pastes raw review text with ratings and confidence scores. Splits each review into individual concerns; classifies every concern as misunderstanding vs real flaw vs requested experiment (plus clarification and disagreement); scores severity x response effort; and produces a prioritized response strategy with per-review character or word budgets matched to the venue's rebuttal format (10k-char OpenReview threads, CVPR one-page PDF, journal revise-and-resubmit). Deterministic parsing and matrix rendering run in bundled stdlib Python scripts. Hands off to write-rebuttal for drafting; treats review text as confidential and never submits anything.
3 other projects use a version of this.
Adopted by
Most recently taken up by
- digitaltracer/agent-skillstheir own version, first seen Nov 15, 2026
- stickerdaniel/linkedin-mcp-servertheir own version, first seen Sep 7, 2026
- tr4m0ryp/linkedin-mcp-servertheir own version, first seen Jun 10, 2026
Other versions3 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 3
- stickerdaniel/linkedin-mcp-server★ 4.2ksame name
- digitaltracer/agent-skills★ 4same name
- tr4m0ryp/linkedin-mcp-server★ 2same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository43 other files it gives its agents.
- docs/llms-full.txtllms.txt
- docs/llms.txtllms.txt
- add-venue-profileSkill
- anonymize-paperSkill
- assess-paperSkill
- benchmark-paperSkill
- check-originalitySkill
- draft-related-workSkill
- draft-surveySkill
- fetch-paperSkill
- find-papersSkill
- fit-page-limitSkill
- literature-reviewSkill
- make-posterSkill
- make-slidesSkill
- match-styleSkill
- orchestrate-paperSkill
- paper-profileSkill
- parse-cfpSkill
- plan-submissionSkill
- polish-proseSkill
- polish-tables-figuresSkill
- preflight-checkSkill
- prepare-artifactsSkill
- prepare-camera-readySkill
- refactor-research-codeSkill
- refactor-structureSkill
- reflect-and-improveSkill
- reflect-paperSkill
- rehearse-qaSkill
- render-workspace-htmlSkill
- select-venueSkill
- simulate-reviewersSkill
- study-exemplarsSkill
- tailor-to-venueSkill
- test-research-codeSkill
- verify-citationsSkill
- verify-claimsSkill
- verify-resultsSkill
- work-with-overleafSkill
- write-abstractSkill
- write-rebuttalSkill
- write-talk-scriptSkill
The file
We split this skill into the rules it gives an agent
6 rules. 0 are common, 2 are distinctive, and 4 are very specific to this project.
0
Common
Most of the 3 edited versions we read keep these.
2
Distinctive
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 23
Always before write-rebuttal.
- Very specific: line 68
Read review.rebuttal_format, review.rebuttal_limit, and deadlines.rebuttal_end from the venue profile.
- Distinctive: line 91
Render the matrix: Run:
- Distinctive: line 137
Never fabricate or soften reviewer text — concerns in the matrix must be traceable to the raw reviews.
- Very specific: line 139
Never claim misunderstanding without a verified anchor in the paper.
- Very specific: line 146
Never submit a response to any system on the user's behalf.
“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Triage Reviews
Convert raw reviewer text into a structured decision artifact: every concern
isolated, classified (misunderstanding / real flaw / requested experiment /
clarification / disagreement), scored severity x effort, and ordered into a
response plan that fits the venue's rebuttal budget. This is the planning
step that runs between "reviews arrived" and write-rebuttal — rebuttals
written straight from raw reviews bury the score-moving points under typo
acknowledgments.
When to use
- "My NeurIPS/ICML/SIGSPATIAL/... reviews are in — help me respond"
- "Triage these reviews" / "what do I address first in my rebuttal?"
- "Reviewer 2 says X but the paper already covers it — how do I handle this?"
- Reviews pasted from OpenReview, EasyChair, CMT, HotCRP, PCS, or a notification email
- Always before
write-rebuttal; also useful for journal revise-and-resubmit responses
Inputs
- Raw review text in a file (e.g.
reviews.txt): the user pastes or exports it from the submission system. Per-platform copy-out instructions and gotchas: references/platform-formats.md. - Venue profile (optional but recommended):
venues/conferences/<venue>-<year>.yml(schema invenues/schema.yml) — suppliesreview.rebuttal_format,review.rebuttal_limit, anddeadlines.rebuttal_end. If missing, runparse-cfpfirst or proceed platform-generic. - The submitted paper (
.tex/PDF, optional): needed to verify misunderstanding claims and fill evidence anchors.
Process
- Stage the raw text — confidentially. Have the user save the reviews
to a local file outside any git repository (or add it to
.gitignore). Review text is confidential at most venues: process it transiently and never commit it. Parse deterministically. Run:
python3 scripts/parse_reviews.py reviews.txt -o triage.jsonAuto-detects the platform; force with
--format openreview|easychair|cmt|hotcrpand catch terse one-liners with--min-words 3if needed. Output is a JSON skeleton: reviewers, scores, canonical sections, and per-concern entries (R1.1,R1.2, ...) withclassification/severity/effortleft null. Exit codes: 0 ok, 1 nothing detected, 2 bad input.- Verify the parse against the raw text. Confirm the reviewer count,
that every weakness/question in the raw text appears as a concern, and
that scores were captured. If a reviewer or concern was missed, fix the
text (insert a
Review Nbanner line) and re-run, or add the concern to the JSON by hand — never silently drop a reviewer point. Recovery steps: references/platform-formats.md, last section. - Resolve the venue's rebuttal mechanics — then re-verify them live.
Read
review.rebuttal_format,review.rebuttal_limit, anddeadlines.rebuttal_endfrom the venue profile. Profiles are a starting point, never ground truth: re-verify the rebuttal format, the character/page limit, whether new experimental results are allowed, and the deadline against the livecfp_url(and the venue's author guidelines) before the user relies on them. State what was verified and when. Ifrebuttal_format: none, say so — triage still guides the camera-ready revision or the next submission. - Classify every concern. Fill
classification,severity,effort,evidence_anchor, and a one-lineresponse_strategyfor each concern intriage.json, applying the decision tree and definitions in references/triage-rubric.md. Rules that bind:- Claim
misunderstandingonly after locating the refuting text in the actual paper — cite section/line inevidence_anchor. - Severity measures threat to acceptance, not reviewer tone; concerns raised by 2+ reviewers escalate one level.
- Effort measures cost to respond within the rebuttal window, not cost to fix the paper.
- Walk the user through any concern where you are uncertain; the user knows the paper.
- Claim
Render the matrix. Run:
python3 scripts/build_matrix.py triage.json --budget <limit> --budget-unit charswith the budget from the verified rebuttal limit (e.g.
--budget 10000 --budget-unit charsfor NeurIPS-style OpenReview;--budget 800 --budget-unit wordsfor a CVPR one-page PDF; omit--budgetfor journal R&R). The script validates the enums (exit 1 with a list of unfilled concerns until classification is complete), computes priority scores and must/should/brief bands, and emits the reviewer summary, severity x effort grid, concern matrix, response plan, and budget allocation (--format jsonfor machine-readable output).- Present and hand off. Walk the user through the must-address band
first, flag any requested experiment that cannot finish before
rebuttal_end, and confirm the strategy lines. Then hand the matrix towrite-rebuttalfor drafting in the venue's format. If new experiments will be run, remind the user to check the venue's policy on new results in rebuttals (step 4) before promising them.
Output
triage.json— structured, classified concern data (machine-readable, reusable bywrite-rebuttal).- A triage matrix report (markdown, via
build_matrix.py): per-reviewer summary with scores, severity x effort grid, priority-ordered concern matrix, must/should/brief response plan, and per-concern budget allocations.
Both stay local; neither should ever contain text the user needs to keep out of a repo — treat them as confidential working files.
Adapt to your discipline
The parser targets CS submission systems. For journal-centric fields, paste
ScholarOne/Editorial Manager reviewer comments, add Reviewer #N banner
lines, and parse with --format cmt; swap the budget step for an unbounded
response-letter plan. The rubric (misunderstanding / real flaw / requested
experiment) is discipline-agnostic.
Guardrails
- Review text is confidential: process transiently, never commit it, never quote it in public artifacts (issues, examples, showcase files).
- Never fabricate or soften reviewer text — concerns in the matrix must be traceable to the raw reviews; quote at most a sentence at a time.
- Never claim
misunderstandingwithout a verified anchor in the paper; honesty rules in references/triage-rubric.md are binding. - Any citation added while planning responses goes through
verify-citationsbefore it reaches a rebuttal. - Re-verify rebuttal format, limits, and deadline against the live
cfp_urlbefore the user relies on them (step 4 is not optional). - Never submit a response to any system on the user's behalf; stop at the
plan (and at the draft, in
write-rebuttal).
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

