任务仍然偏 synthetic / templated
- Stable ID:
codex-user-prompt:1770ed9eefb4da1c
- Source kind:
codex-session-user
- Category:
general-high-quality-prompt
- Timestamp:
2026-05-10T10:53:18.732Z
- Semantic hash:
1770ed9eefb4da1cc962fb66de3b828cf0fd17047a463e6e79471402219a50b2
- Public handling: selected full-text prompt with secret filtering.
Prompt Text
任务仍然偏 synthetic / templated
这不是致命伤，因为很多 benchmark 都有程序化生成部分，但必须承认 DoneBench 是 controlled semantic benchmark，不是现实 UI benchmark。
那怎么样能够加个现实的那种
human calibration我朋友打算做，但是他刚看，0基础，怎么完整的跟他交代
Reuse Notes
- EXTRACTED: This is a selected Codex prompt or automation prompt from the local Codex corpus.
- INFERRED: Future agents can reuse its structure, constraints, and acceptance criteria when creating similar Codex workflows.