Skip to content

sidiangongyuan/codex-skills-library / skills/paper-review-panel

paper-review-panel

Verified

Use when independently reviewing a research-paper draft before submission, running a mock top-conference panel, assessing readiness or accept/reject risk, or predicting reviewer concerns for venues such as CVPR, ICCV, ECCV, ICLR, NeurIPS, and AAAI. Does not draft or audit author responses after official reviews arrive.

FileKindSize
LICENSEother1.0 KB
SKILL.mdskill md5.7 KB
agents/openai.yamlconfig296 B
references/output-format.mdmarkdown2.4 KB
references/review-checklist.mdmarkdown4.4 KB
references/reviewer-roles.mdmarkdown3.8 KB

references/review-checklist.md

View on GitHub
# Review Checklist

Use this checklist for deep reviews, readiness audits, or when the paper is
close to submission.

## Contribution And Positioning

- Is the central problem clear by the end of the introduction?
- Is the paper's contribution specific rather than a list of components?
- Does related work make the gap obvious without strawman comparisons?
- Is the claimed novelty believable for the target venue?
- Are limitations acknowledged without weakening the main claim unnecessarily?

## Related Work And Novelty Risk

- For high-risk reviews, was a recent-literature audit run instead of relying
  only on memory or the paper's current bibliography?
- Were direct and adjacent recent works searched with terminology variants,
  older names, acronyms, and neighboring task expressions?
- Are close works classified as direct competitor, claim limiter, table
  candidate, prose citation, background, or irrelevant?
- Does the paper avoid first/SOTA/only-new-benchmark language when adjacent work
  limits that claim?
- Are missing or ambiguous citations framed as reviewer risk with source
  coverage limits?
- If source coverage is partial, is the score or acceptance-risk judgment
  explicitly conditional on that limitation?

## Claims And Evidence

- Can a reviewer identify one central claim and its primary evidence before
  encountering ablations, reliability studies, or diagnostics?
- Does every main claim point to a table, figure, experiment, theorem, dataset
  statistic, annotation audit, or citation?
- Are negative or mixed results framed honestly?
- Are appendix diagnostics used as diagnostics rather than main proof?
- Are deployment, robustness, calibration, generalization, or causality claims
  supported narrowly enough?

## Method And Protocol

- Is the task definition explicit?
- Are inputs, outputs, supervision, and inference-time information legal?
- Are train and evaluation protocols aligned?
- Are preprocessing, coordinate systems, prompts, model checkpoints, and
  conversion steps described enough to avoid ambiguity?
- Is there any hidden leakage from labels, future information, target actions,
  or evaluation artifacts?

## Experiments And Metrics

- Are baselines fair and clearly labeled?
- Are metrics defined by measured quantity, aggregation population,
  directionality, and units?
- Does every delta name its reference and make the sign convention clear?
- Are sample counts, splits, and checkpoint policies consistent?
- Are ablations tied to the main mechanism?
- Are confidence intervals, paired comparisons, or stratified analyses used
  when small deltas drive the conclusion?
- Are command/class imbalance and metric disagreement handled explicitly?
- Is repeated-seed evidence requested only when variance could change the main
  claim, rather than mechanically for an expensive training task?
- If the draft uses `--`, are those cells treated only as planned structure,
  with no result claim inferred from them?

## Figures And Tables

- Can each figure/table be understood without reading long surrounding prose?
- Does each main table answer one reviewer question, with secondary diagnostics
  clearly subordinate to the main comparison?
- Are captions concise and claim-aligned?
- Are colors, arrows, deltas, and highlights semantically consistent?
- Are table values traceable to artifacts?
- Are plot axes, units, legends, and labels consistent?
- Are PDF-rendered figures/tables unclipped and readable?

## Writing And Structure

- Does the abstract state problem, gap, contribution, and evidence clearly?
- Does each section have one job?
- Are terms and symbols introduced once and used consistently?
- Are paragraphs cohesive rather than run-log fragments?
- Are citations present for factual comparisons and related work claims?
- Are duplicate aliases, opaque coined metrics, and unnecessary diagnostic
  concepts removed before they make the paper tiring to read?
- Is paper-facing text free of author instructions and internal labels such as
  `TODO`, `TBD`, or `pending`?

## Reproducibility And Appendix

- Are released artifacts, review scope, and dataset terms stated precisely?
- Do appendix tables agree with main-paper numbers?
- Are qualitative examples representative and not overclaimed?
- Are draft markers, internal logs, and implementation notes removed from
  paper-facing text?
- Is the appendix useful as evidence and reproducibility support, not a hidden
  rebuttal?
rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first.Stars: 194kUpdated Claude CodeCodexVerifiedInstall