[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M69B75RBJCMTN4GVY7T0QP30":3,"skill-same:01M69B75RBJCMTN4GVY7T0QP30":82},{"skillNo":4,"name":5,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":17,"publishedAt":18,"linkOnly":19,"markdownContent":20,"markdownBody":20,"frontMatter":21,"refName":22,"commitSha":23,"skillDirectory":24,"forksCount":25,"repositoryHtmlUrl":26,"repositoryHomepageUrl":27,"ownerHtmlUrl":28,"ownerBlogUrl":29,"ownerDescription":30,"updatedAt":31,"install":32,"resources":45},"01M69B75RBJCMTN4GVY7T0QP30","core",null,"microsoft\u002FRD-Agent","microsoft","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F6154722?v=4","01M698VYDXZ6JW5WH3CRV650R8","Microsoft","01M69B72SSTNMTG8464HQG4FAY",15361,"MIT","https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FRD-Agent\u002Fblob\u002Fmain\u002Frdagent\u002Fscenarios\u002Frl\u002Fautorl_bench\u002Fcore\u002Fskill.md",[],[5],"2026-11-10 10:12:30",false,"你负责维护实验的累积运行总结文件 `reports\u002Fsummary.md`。\n\n每轮实验结束后，你需要用 file_editor 在 `reports\u002Fsummary.md` **末尾追加**一个新 section。\n\n如果 `reports\u002Fsummary.md` 不存在，先创建文件，首行写 `# 运行总结`。\n\n每轮追加的 section 格式（严格遵守，不增删字段）：\n\n```\n## Iteration N (YYYY-MM-DD HH:MM, 耗时 Xs)\n- **状态**: ✅ 成功 \u002F ❌ 失败 (exit_code=X)\n- **Score**: X | Improvement: X | Best: X (iter N)\n- **训练类型**: GRPO \u002F SFT \u002F PPO \u002F copy_model \u002F placeholder \u002F unknown\n- **关键配置**: lr=X, epochs=X, batch=X, ...（从代码中提取）\n- **做了什么**: 具体策略（训练方法、reward 函数设计、数据处理等）\n- **为什么**: 为什么选择这个策略（基于上轮结果的推理）\n- **问题\u002F进步**: 发现了什么问题，或相比上轮取得了什么进步\n- **关键代码**: 最能体现本轮策略的 3-5 行代码\n- **代码片段（上下文）**: 从 train.py 复制 15-40 行可运行上下文，标注函数名\u002F行号范围；若与上轮相同写“与 Iteration N 相同，无变更”\n```\n\n数据来源建议（按优先级）：\n- Score\u002FImprovement\u002FBest: scores.json 或服务器返回\n- 状态\u002F耗时\u002Fexit_code: run.log\n- 训练类型\u002F关键配置\u002F关键代码: code\u002Ftrain.py\n- 失败根因: agent.log + run.log\n\n规则：\n1. **追加**，不要覆盖已有内容\n2. 必须按 Iteration 递增写入：读取 `reports\u002Fsummary.md` 中最后一个迭代号，当前必须是 N+1；若不满足先修正再写\n3. 分析 code下面的 源码，提取训练类型和超参数；若无法确定写 unknown，不留空\n4. 如果训练失败，必须给出可定位的根因（日志片段或错误类型），并标注 failure_type（如：code_error_runtime \u002F rollout_logic_wrong \u002F timeout_no_submission \u002F copy_model_fallback \u002F training_diverged \u002F unknown）\n5. “做了什么”“为什么”是最重要字段，必须可复现、可检验，且“为什么”必须引用上轮证据（scores.json\u002Frun.log\u002Fagent.log）\n6. **问题\u002F进步** 必须包含过程指标或失败类型（如：valid_submission_rate \u002F first_valid_idx \u002F time_to_first_improvement \u002F time_used_ratio \u002F failure_type）\n7. 长代码片段每轮最多 1 段，最多 40 行；优先贴本轮新增或改动处；若与上轮相同写“与 Iteration N 相同，无变更”\n8. 若与上一轮根因相同，避免整段重复，明确写出“新增证据\u002F新增尝试\u002F无新增”\n9. 同步追加 `reports\u002Fsummary.jsonl`（同目录），每轮一行 JSON，字段建议：\n   - iteration, timestamp, duration_s, status, exit_code\n   - score, improvement\n   - train_type, failure_type\n   - 仅保留可量化字段（图表使用），不写 why\u002Fwhat\u002Fnext_step 等长文本\n   - metrics 由 benchmark 后处理统一计算（不在 jsonl 中填写）\n   - 示例：\n     {\"iteration\": 3, \"timestamp\": \"2026-03-10 12:34\", \"duration_s\": 812, \"status\": \"success\", \"exit_code\": 0, \"score\": 65.0, \"improvement\": 20.0, \"train_type\": \"GRPO\", \"failure_type\": \"unknown\"}\n10. 写入后自检 section 完整性：上述字段必须齐全，缺失则补全后再结束\n",{},"main","484776c211e4fbbeef03e0ec00d6bbee7362a4f4","rdagent\u002Fscenarios\u002Frl\u002Fautorl_bench\u002Fcore",2050,"https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FRD-Agent","https:\u002F\u002Frdagent.azurewebsites.net\u002F","https:\u002F\u002Fgithub.com\u002Fmicrosoft","https:\u002F\u002Fopensource.microsoft.com","Open source projects and samples from Microsoft","2026-10-22 00:01:22",{"source":7,"skillName":5,"commands":33,"prompt":43,"zipAvailable":44},[34,37,40],{"tool":35,"command":36},"npx","npx skills add microsoft\u002FRD-Agent --skill core",{"tool":38,"command":39},"bunx","bunx skills add microsoft\u002FRD-Agent --skill core",{"tool":41,"command":42},"pnpm","pnpm dlx skills add microsoft\u002FRD-Agent --skill core","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FRD-Agent\u002Fblob\u002Fmain\u002Frdagent\u002Fscenarios\u002Frl\u002Fautorl_bench\u002Fcore\u002Fskill.md",true,[46,52,56,62,66,70,74,78],{"path":47,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":51},"__init__.py","SCRIPT","脚本",2566,"https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FRD-Agent\u002Fblob\u002Fmain\u002Frdagent\u002Fscenarios\u002Frl\u002Fautorl_bench\u002Fcore\u002F__init__.py",{"path":53,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":54,"sourceHtmlUrl":55},"evaluator.py",5436,"https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FRD-Agent\u002Fblob\u002Fmain\u002Frdagent\u002Fscenarios\u002Frl\u002Fautorl_bench\u002Fcore\u002Fevaluator.py",{"path":57,"resourceType":58,"resourceTypeDesc":59,"sizeBytes":60,"sourceHtmlUrl":61},"instructions.md","REFERENCE","参考文档",5919,"https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FRD-Agent\u002Fblob\u002Fmain\u002Frdagent\u002Fscenarios\u002Frl\u002Fautorl_bench\u002Fcore\u002Finstructions.md",{"path":63,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":64,"sourceHtmlUrl":65},"metrics.py",5063,"https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FRD-Agent\u002Fblob\u002Fmain\u002Frdagent\u002Fscenarios\u002Frl\u002Fautorl_bench\u002Fcore\u002Fmetrics.py",{"path":67,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":68,"sourceHtmlUrl":69},"opencompass.py",8518,"https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FRD-Agent\u002Fblob\u002Fmain\u002Frdagent\u002Fscenarios\u002Frl\u002Fautorl_bench\u002Fcore\u002Fopencompass.py",{"path":71,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":72,"sourceHtmlUrl":73},"server.py",14783,"https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FRD-Agent\u002Fblob\u002Fmain\u002Frdagent\u002Fscenarios\u002Frl\u002Fautorl_bench\u002Fcore\u002Fserver.py",{"path":75,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":76,"sourceHtmlUrl":77},"ui.py",4650,"https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FRD-Agent\u002Fblob\u002Fmain\u002Frdagent\u002Fscenarios\u002Frl\u002Fautorl_bench\u002Fcore\u002Fui.py",{"path":79,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":80,"sourceHtmlUrl":81},"utils.py",12735,"https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FRD-Agent\u002Fblob\u002Fmain\u002Frdagent\u002Fscenarios\u002Frl\u002Fautorl_bench\u002Fcore\u002Futils.py",[]]