{
 "selected": [
  {
   "cid": "ai_session_search_archive#0"
  },
  {
   "cid": "compressor#0"
  },
  {
   "cid": "deployment_github_actions_koyeb#0"
  },
  {
   "cid": "mac_universal_clipboard#0"
  },
  {
   "cid": "openreview#1"
  },
  {
   "cid": "workflow_bilibili_whisper_transcription#1"
  },
  {
   "cid": "workflow_watchdog#0"
  },
  {
   "cid": "workflow_parallel_subagents#1"
  },
  {
   "cid": "codex_cli#1",
   "edits": {
    "rubric": {
     "must_include": [
      "写成配置顶层的裸键 web_search = \"disabled\"",
      "值必须是字符串 \"disabled\"，布尔值 false 等会被拒绝"
     ],
     "reference_answer": "在基础配置（或该次加载的 profile）顶层写 web_search = \"disabled\"；只能用字符串 \"disabled\"，false/off 等会被拒绝。"
    }
   },
   "note": "dropped the base-vs-profile rubric item; question asks level and value"
  },
  {
   "cid": "ai_session_profanity_rate#1"
  },
  {
   "cid": "apple_photos#1"
  },
  {
   "cid": "asr#1",
   "edits": {
    "rubric": {
     "must_include": [
      "单个文件超过 7200 秒（2 小时）会被拒绝，所以 2 小时 10 分钟的单文件不能直接处理",
      "在模型加载之前就被拒绝"
     ],
     "reference_answer": "不能直接处理：单文件超过 7200 秒会在加载模型之前被拒绝；切成多个文件可以分别处理（但说话人编号不会跨文件延续）。"
    }
   },
   "note": "rubric trimmed to the two facts the excerpt states; splitting is allowed but not required for credit"
  },
  {
   "cid": "ct_education#0"
  },
  {
   "cid": "dataforseo#0"
  },
  {
   "cid": "eink_diary#1"
  },
  {
   "cid": "skill_google_maps_routing#0"
  },
  {
   "cid": "hybrid_ai_video#1",
   "edits": {
    "rubric": {
     "must_include": [
      "不能直接再发一次创建请求：结果不确定的提交必须阻止自动重提，不要用新的 POST 去续上任务"
     ],
     "reference_answer": "不能；结果不确定的提交要阻止自动重新提交，也不要用新的 POST 去恢复任务，应先核实（查询）任务状态。"
    }
   },
   "note": "rubric reduced to the core rule"
  },
  {
   "cid": "ecg_plot#1"
  },
  {
   "cid": "koyeb#0"
  },
  {
   "cid": "narwal#1"
  },
  {
   "cid": "skill_open_router_data_scraper#0"
  },
  {
   "cid": "skill_opencode_data#0"
  },
  {
   "cid": "video-scoring-audio#1"
  },
  {
   "cid": "speaker_notes#0"
  },
  {
   "cid": "skill_tavily#1",
   "edits": {
    "question": "我想把 Tavily 命令行 search 的完整结果直接打印在控制台，加个 --stdout 可以吗？如果加了会不会真正发起请求，退出码是多少？"
   },
   "note": "question now names Tavily (the drafted one named no tool); skill_firecrawl states the same rule for its own CLI, so opening it can also yield a correct answer"
  },
  {
   "cid": "zun_music#1"
  },
  {
   "cid": "skill_image_generation#0"
  },
  {
   "cid": "skill_process_launcher#1",
   "edits": {
    "rubric": {
     "must_include": [
      "要并存：用同样的请求重新提交并带 \"confirm_conflict\": true",
      "要替换：先用 POST /scheduled/{id}/cancel 取消冲突的旧任务再提交"
     ],
     "reference_answer": "409 时什么都没创建；要两条并存就带 confirm_conflict: true 重新提交；要替换就先 cancel 旧任务再提交新任务。"
    }
   },
   "note": "dropped the exact-time-only rubric item, which the question does not ask"
  },
  {
   "cid": "skill_resend_email#1"
  },
  {
   "cid": "skill_typefully_twitter#1"
  }
 ],
 "rules": "Kept only candidates with 0/3 no-skill baseline correct (stricter than the <2/3 rule), one task per target skill, spread across 8 context-infrastructure skills and 22 skills from 22 different ecosystem repos. Dropped candidates whose key fact is stated in another pool skill (voice-lora vs voice-lora-rewrite, external_prose_lint vs writing_workflows, ios_development vs simulator_testing, m5stack_sticks3_test_loop vs m5stack_sticks3_esp_idf, skill_firecrawl vs skill_tavily kept only tavily), candidates relying on placeholder values (growth_analytics#0 example.com) or private paths (1password#1, delayed_execution#0), and vague questions without a domain anchor (share_report#0)."
}