用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt;运行前请先检查来源。
npx skills add https://github.com/expectedparrot/katz --skill review-figures命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地?可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
基于 SOC 职业分类
| name | review-figures |
| description | Send paper figures to vision-capable models for feedback on clarity, design, and presentation |
| allowed-tools | Read, Bash, Glob, Grep |
| user-invocable | true |
Sends each figure from the paper to vision-capable models via EDSL for feedback on clarity, labeling, design quality, and caption adequacy. Results are recorded as eval responses under the "figures" category.
/review-figures
.katz/versions/<commit>/paper/.
(This happens automatically when registering with katz paper register — sibling image files
are copied alongside the manuscript.)edsl must be installed.Run katz paper status to confirm registration. Check that images exist:
ls .katz/versions/$(katz paper status | python3 -c "import sys,json; print(json.load(sys.stdin)['data']['commit'])")/paper/*.png
# All figures, 2 models (Claude Opus + GPT-5.4)
python <katz-skills-path>/review-figures/scripts/edsl_review_figures.py
# Dry run to see what would be sent
python <katz-skills-path>/review-figures/scripts/edsl_review_figures.py --dry-run
# Single model
python <katz-skills-path>/review-figures/scripts/edsl_review_figures.py --models 1
The script:
Results appear in the HTML report under the Evaluations section (category: "figures"), or via:
katz eval results --category figures
Each figure is assessed on five dimensions: