pengjunchi0/codex-visio-paper-figure-skill: does the agent reach for it? 2026-10-11
Twelve Codex skills were installed together and each agent was sent three requests meant for this skill (visio-image-rebuilder), in a throwaway sandbox. A run stopped as soon as a skill was opened.
Then the same requests with 100 and 500 other real skills installed beside the twelve. Codex ran gpt-6-astra, Claude Code ran Claude Opus 5.5.
All levels · The requests · How a run is scored
At 500, Codex opened image-to-editable-ppt first on one request, then searched the disk and read this skill too; we count the first one opened.
| Request | Codex, 0 others | Codex, 100 others | Codex, 500 others | Claude Code, 0 others | Claude Code, 100 others | Claude Code, 500 others |
|---|---|---|---|---|---|---|
| asked plainly Recreate the attached reference image as an editable diagram, preserving separate shapes, connectors, labels, and panel alignment. Export it as an editable .vsdx plus PNG, SVG, PDF, and PPTX; reference: a three-panel signaling pathway with blue receptors, orange arrows, and green response nodes. | opened it 7s | opened it 8s | no skill opened time limit | opened it 4s | opened it 3s | opened it 3s |
| described without the skill's words Turn this reference into a fully editable vector diagram rather than a flattened picture, matching its layout, colors, typography, and scientific notation, and provide the editable source plus PNG, SVG, PDF, and PPTX exports. Reference: a three-panel signaling pathway with blue receptors, orange arrows, and green response nodes. | opened it 8s | opened it 6s | opened image-to-editable-ppt | opened it 5s | opened it 3s | opened it 4s |
| in the other language 请根据下面的参考图重建一个可编辑的图表，保留独立形状、连接线、标签和面板对齐关系。请导出可编辑的 .vsdx 文件，以及 PNG、SVG、PDF 和 PPTX；参考图内容：一个包含蓝色受体、橙色箭头和绿色响应节点的三面板信号通路图。 | opened it 8s | opened it 11s | no skill opened time limit | opened it 4s | opened it 3s | opened it 14s |