只读研究任务，不要修改文件。请浏览/查阅真实任务/leaderboard/reproducibility 型顶会项目：SWE-bench、SWE-benc…
- Stable ID:
codex-user-prompt:39e440060264ca1f
- Source kind:
codex-session-user
- Category:
coding-agent-workflow
- Timestamp:
2026-05-10T09:09:39.534Z
- Semantic hash:
39e440060264ca1fabc4944586560c39290ea4e4dab02e2c1d3f616241be4bad
- Public handling: selected full-text prompt with secret filtering.
Prompt Text
只读研究任务，不要修改文件。请浏览/查阅真实任务/leaderboard/reproducibility 型顶会项目：SWE-bench、SWE-bench Verified、MLE-bench、GAIA、BrowseComp/SpreadsheetBench/OfficeBench 或类似 agent/task benchmarks。目标是提炼数据来源、验证集、人工校准、artifact release、leaderboard 防污染、cost/latency/reporting 标准。输出：1) DoneBench 目前缺哪些顶会级 artifact 模块；2) 哪些模块有必要补、哪些不值得；3) 是否需要新实验还是写作/文档就够；4) 具体下一步计划。请附来源链接。
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.