{"id":51285904,"url":"https://github.com/binary-husky/team_bench_gen","last_synced_at":"2026-06-30T05:32:36.337Z","repository":{"id":365655450,"uuid":"1270973997","full_name":"binary-husky/team_bench_gen","owner":"binary-husky","description":null,"archived":false,"fork":false,"pushed_at":"2026-06-28T18:24:15.000Z","size":29808,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-06-28T20:15:01.781Z","etag":null,"topics":[],"latest_commit_sha":null,"homepage":null,"language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/binary-husky.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2026-06-16T08:06:01.000Z","updated_at":"2026-06-28T18:24:21.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/binary-husky/team_bench_gen","commit_stats":null,"previous_names":["binary-husky/team_bench_gen"],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/binary-husky/team_bench_gen","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/binary-husky%2Fteam_bench_gen","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/binary-husky%2Fteam_bench_gen/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/binary-husky%2Fteam_bench_gen/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/binary-husky%2Fteam_bench_gen/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/binary-husky","download_url":"https://codeload.github.com/binary-husky/team_bench_gen/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/binary-husky%2Fteam_bench_gen/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":34954283,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-06-30T02:00:05.919Z","response_time":92,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":[],"created_at":"2026-06-30T05:32:35.602Z","updated_at":"2026-06-30T05:32:36.330Z","avatar_url":"https://github.com/binary-husky.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# team_bench_gen\n\n一个**可验证研究课题**的批量生成与沉淀仓库。每个课题都是一份可直接派发给执行 Agent 和 Judge Agent 的任务单：执行 Agent 读材料 / 做实验 / 写结论，Judge Agent 只看结论文件给通过/不通过。\n\n生成者**只负责拟写课题**，不运行实验、不写实验结论、不替后续 Agent 做验证。\n\n---\n\n## 仓库结构\n\n```\nteam_bench_gen/\n├── generated_task_await_verify/         # 已生成、尚未通过验证的课题\n│   ├── cec/      cec_{01..08}.md  + cec_material/\n│   ├── hpob/     hpob_{01..05}.md + hpob_material/\n│   └── {new_big_subject}/               # 新大课题落到这里\n│\n├── generated_task_verified/             # 已通过验证的课题（可作为拓展模板）\n│   ├── muon/     muon_{01..09}.md + muon_material/\n│   └── {new_big_subject}/\n│\n└── generated_task_verified_experiments/ # 已验证课题的实验产物（summary、log 等）\n    ├── muon/\n    └── {new_big_subject}/\n```\n\n**目录约定**：一个大课题 (big_subject) 一个子目录；该大课题下所有小课题共享同一个 material 目录，material 不嫌多。\n\n---\n\n## 课题文件格式\n\n每个 `*.md` 必须使用以下结构：\n\n```text\n[Agents]\n\n读给定材料，做实验，写结论。\n\n\u003c具体实验任务、变量、对照组、训练/分析长度、记录指标、输出文件\u003e\n\n---\n\n[Judge]\n\nLook at `./summary_\u003ctopic\u003e.md`, check whether conclusion cover the following points (最多不超过 3 点，每点都必须清晰、可验证)\n\n1. \u003c结果判定点 1\u003e\n2. \u003c结果判定点 2\u003e\n3. \u003c结果判定点 3\u003e\n```\n\n### `[Agents]` 写法\n\n`[Agents]` 是给执行 Agent 的任务布置，只显式要求三件事：\n\n1. 读给定材料。\n2. 做实验或分析。\n3. 写结论。\n\n不要在任务文本里显式要求 Agent 查找资料、补充资料、联网搜索或登记材料。如果 Agent 自发查资料，那是 Agent 自己的行为，不应写成任务要求。\n\n### 给定材料写法\n\n使用\"给定材料\"这个统称，**不要**显式提及材料的具体路径、文件名、项目内已有说明文件、既有总结、其他研究记录或其他执行者的结论。\n\n| ✅ 推荐 | ❌ 不推荐 |\n|---|---|\n| `读给定材料` | `读某个具体材料文件` |\n|  | `参考项目内已有说明文件` |\n|  | `根据已有总结` |\n|  | `查看其他研究记录中的结论` |\n\n### `[Judge]` 写法\n\n`[Judge]` 是**研究结果判定条件**，不是实验执行说明。\n\n- 固定实验设置（epoch 数、优化器、学习率等）写在 `[Agents]`，不要写成 Judge 判定条件。\n- Judge 条件只检查研究结果和结论，**最多 3 点**，每点都必须清晰、可验证。\n\n### 输出文件写法\n\n每个课题指定**一个**结果文件，通常命名为 `./summary_\u003ctopic\u003e.md`。Judge 只查看该文件进行判定。**不要**要求 Agent 修改已有总报告、已有说明文件、已有总结或其他已有研究记录。\n\n---\n\n## Skill 1：从论文产生新大课题\n\n从一个论文 PDF 抽取一个大课题 (big_subject)，围绕它生成 **5 个**可验证小课题 (small_subject)，复杂度和难度缓慢提升。\n\n**示例参考论文**：\u003chttps://arxiv.org/pdf/2106.06257\u003e\n\n### 流程\n\n1. 在 `generated_task_await_verify/{big_subject}/` 下新建子目录（暂时**不允许**创建全新大课题——如需新建请先确认）。\n2. 把原始材料复制到 `generated_task_await_verify/{big_subject}/{big_subject}_material/`。**材料必须是直接从互联网获取的、未二次加工的原始材料**（论文 PDF、官方仓库源码、作者博客原文等）。\n3. 写 5 个小课题：`{big_subject}_{01..05}.md`，遵循上面的格式。\n\n### 5 个小课题的硬约束\n\n| 编号 | 类型 | 要求 |\n|---|---|---|\n| **01** | 逻辑推理题（无实验） | 通过逻辑推理得到答案，**原文不得存在直接答案**；**只能有一个问题**；生成者必须给出**标准答案**，设计 **1** 个评价维度 |\n| **02–05** | 实验题 | 需要做实验得出结论（参考 `generated_task_verified/muon/muon_02.md`）；生成者不亲自做实验，只做合理猜想，设计 **2–3** 个评价维度 |\n\n通用约束：\n- 不得创建过于相似的小课题。\n- 每个课题都要求解题者输出一个 `summary_xxx.md` 作为结论。\n\n---\n\n## Skill 2：从已有小课题拓展更多小课题\n\n从一个**已验证**的小课题出发，派生出更多同主题、不同角度的小课题。\n\n**示例**：\n\n```\n原始：generated_task_verified/muon/muon_02.md\n拓展：generated_task_verified/muon/muon_03.md\n      generated_task_verified/muon/muon_04.md\n      generated_task_verified/muon/muon_05.md\n      generated_task_verified/muon/muon_06.md\n```\n\n### 流程\n\n1. 拓展出的新课题**必须**落到 `generated_task_await_verify/{big_subject}/`（因为没验证）。\n2. 如果需要新材料，复制到 `generated_task_await_verify/{big_subject}/{big_subject}_material/`，沿用原始材料规则。\n3. 命名沿用 `{big_subject}_{下一个编号}.md`。\n\n### 课题选择原则\n\n- 课题必须**可验证**——Judge 能基于结论文件给出明确的通过/不通过。\n- 拓展课题之间不得过于相似；每个新课题都应有独立的研究问题。\n- 编号连续，新增/删除后保持编号一致。\n\n---\n\n## 命名规则汇总\n\n| 对象 | 路径模板 |\n|---|---|\n| 课题 md | `{big_subject}/{big_subject}_{编号}.md` |\n| 材料文件夹 | `{big_subject}/{big_subject}_material/` |\n| 结果文件 | `./summary_\u003ctopic\u003e.md` |\n\n---\n\n## 自检清单\n\n生成或拓展任务单后，检查以下事项：\n\n1. 每个课题都有 `[Agents]`。\n2. 每个课题都有 `[Judge]`。\n3. `[Agents]` 中只显式要求\"读给定材料，做实验，写结论\"。\n4. 没有显式要求 Agent 查找、补充、登记或下载材料。\n5. 没有显式提及材料文件名、已有说明文件、既有总结或其他研究记录结论。\n6. 固定实验设置写在 `[Agents]`，没有写成 Judge 判定条件。\n7. Judge 条件只检查研究结果和结论，最多 3 点。\n8. 每个课题都有明确输出文件。\n9. 课题编号连续。\n10. 推荐顺序与删除或新增后的编号一致。\n11. `_01` 是逻辑推理题（无实验、单问题、有标准答案、1 个评价维度）。\n12. `_02` 起是实验题（2–3 个评价维度）。\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fbinary-husky%2Fteam_bench_gen","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fbinary-husky%2Fteam_bench_gen","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fbinary-husky%2Fteam_bench_gen/lists"}