joeseesun/qiaomu-goal-meta-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 (qiaomu-goal-meta-skill), 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
| Request | Codex, 0 others | Codex, 100 others | Codex, 500 others | Claude Code, 0 others | Claude Code, 100 others | Claude Code, 500 others |
|---|---|---|---|---|---|---|
| asked plainly Turn this vague request into a strong `/goal` command with a clear outcome, success criteria, verification commands, constraints, boundaries, iteration policy, completion evidence, and pause conditions: “Improve the checkout flow.” | opened it 16s | opened it 6s | no skill opened finished | opened it 3s | opened it 2s | opened it 3s |
| described without the skill's words Rewrite this vague engineering request as a bounded, actionable instruction for an agent, including the desired outcome, measurable success criteria, checks to run, scope limits, iteration rules, proof of completion, and conditions for stopping to ask questions: “Improve the checkout flow.” | opened it 17s | opened it 5s | no skill opened finished | opened it 3s | opened it 4s | opened it 3s |
| in the other language 请把下面这个模糊请求改写成一条完善的 `/goal` 指令，包含明确目标、成功标准、验证命令、约束、范围边界、迭代规则、完成证据以及暂停或受阻条件：“改进结账流程。” | opened it 5s | opened it 18s | no skill opened finished | opened it 3s | opened it 3s | opened it 3s |