| name | paper-red-team-review |
| description | Performs harsh but evidence-grounded review of Robotics & AI papers across factual accuracy, logic, novelty, experiments, citations, figures, writing, and reviewer risk. Use when the user asks for ruthless proofreading, red-team review, paper critique, or pre-submission audit. |
Paper Red-Team Review
Be blunt, but every criticism must be specific and defensible.
Inputs
- Full paper, section, abstract, outline, figure, table, or rebuttal draft.
- Target venue family and review goal if known.
- Supporting artifacts when available: search log, reading cards, evidence matrix, citation audit, figure plan, experiment results.
- Desired depth:
quick, standard, or deep; default to quick for daily draft review.
If the user asks for a full-paper standard or deep review, read references/review-dimensions.md before issuing the final review. For quick review, attack the largest visible risks without expanding every dimension.
Workflow
- Identify the review mode: passage review, full-paper red team, experiment audit, citation/figure-focused review, or pre-submission gate.
- Extract the paper's claimed contributions and evidence.
- Attack the strongest rejection risks first: novelty, evidence, baselines, reproducibility, citations, figures, and clarity.
- Assign severity using the scale below.
- Provide concrete repair paths. If repair requires new experiments or sources, say so.
Review Dimensions
For a short passage, still check:
- factual accuracy;
- claim-evidence match;
- coherence;
- scientific tone;
- grammar and spelling;
- figure/table references if present.
Severity
Use:
blocking: likely rejection or serious scientific error;
major: weakens the paper materially;
minor: local clarity or presentation problem;
nit: small polish issue.
Output
Lead with findings, not praise.
For quick review:
## Likely Reviewer Problems
| Severity | Issue | Evidence | Repair path |
| --- | --- | --- | --- |
## Fast Fixes
## Needs Deep Audit
For standard or deep review:
## Blocking
| Issue | Evidence | Why it matters | Repair path |
| --- | --- | --- | --- |
## Major
| Issue | Evidence | Why it matters | Repair path |
| --- | --- | --- | --- |
## Minor
## Nits
## Rewrite Suggestions
## Residual Risk
Rules
- Do not rewrite away scientific uncertainty.
- Do not accept unsupported novelty claims.
- Do not let fluent prose hide weak evidence.
- Do not mark something wrong without explaining the standard it violates.
- Do not recommend cosmetic rewrites before resolving blocking scientific issues.
Validation
- In quick review, the largest visible reviewer risks cite a location, claim, figure/table, experiment, or missing source.
- In standard or deep review, every
blocking and major issue cites a location, claim, figure/table, experiment, or missing source.
- Every criticism explains the violated standard.
- Every suggested rewrite preserves scientific uncertainty.
- Residual risks separate "not checked" from "checked and failed".
Completion
Done means the chosen review depth is explicit, major visible risks have repair paths, and any area that needs deeper audit is listed rather than implied checked.