← Codex skills

Evianis/travel-photo-abstraction: 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 (travel-photo-abstraction), 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

RequestCodex, 0 othersCodex, 100 othersCodex, 500 othersClaude Code, 0 othersClaude Code, 100 othersClaude Code, 500 others
asked plainly
Analyze the uploaded photograph, preserve it exactly unchanged in the upper panel, and reconstruct its essential visual relationships as a new nonliteral abstraction in a clean lower panel, guided by the 19 provided reference images. Add a number, date, and short phrase to the lower panel.
opened it 15sopened it 7sno skill opened finishedopened it 3sopened it 2sopened it 4s
described without the skill's words
Create a clean two-part editorial image: leave my uploaded travel photo untouched at the top, then make a nonliteral composition below that distills its key colors, forms, textures, and spatial relationships rather than redrawing the scene. Use the 19 supplied visual references as guidance, and include a number, date, and short phrase in the bottom section.
opened it 7sopened it 8sno skill opened time limitopened it 3sopened it 2sopened it 3s
in the other language
请分析我上传的照片,将照片原样完整保留在上方,并参考提供的 19 张图像,在下方重构其核心视觉关系,创作一个全新的非具象抽象构图。请在下方区域加入一个数字、日期和简短短语。
opened it 10sopened it 7sno skill opened finishedopened it 3sopened it 3sopened it 3s