[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6C2Q8X03RTC0Y9X6PRGJC59":3,"skill-same:01M6C2Q8X03RTC0Y9X6PRGJC59":60},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":22,"publishedAt":34,"linkOnly":35,"markdownContent":36,"markdownBody":37,"frontMatter":38,"refName":39,"commitSha":40,"skillDirectory":41,"forksCount":42,"repositoryHtmlUrl":43,"repositoryHomepageUrl":6,"ownerHtmlUrl":44,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":45,"install":46,"resources":59},"01M6C2Q8X03RTC0Y9X6PRGJC59","skeptic-agent",null,"Define and run skeptic exit criteria for non-trivial tasks — independent verification agent with inverted incentive to find gaps","jleechanorg\u002Fjleechan-skills","jleechanorg","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F221404966?v=4","01M6C1VJ99XG950X1DWPEBCBGX","01M6C1VJDK5YVDNT3362RT5WSB",6,"MIT","https:\u002F\u002Fgithub.com\u002Fjleechanorg\u002Fjleechan-skills\u002Fblob\u002Fmain\u002F.claude\u002Fskills_archive\u002F2026-08-27-historical-zero-use\u002Fskeptic-agent\u002FSKILL.md",[17],{"categoryNo":18,"categoryName":19,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000023","Skill与智能体开发","SCENARIO","使用场景",[23,24,25,26,27,28,29,30,31,32,33],"skeptic agent","skeptic","tasks","run","verification","find","exit","define","with","for","gaps","2026-11-10 10:12:30",false,"---\nname: skeptic-agent\ndescription: Define and run skeptic exit criteria for non-trivial tasks — independent verification agent with inverted incentive to find gaps\n---\n\n> **DELETED SYSTEM (2026-07-09).** The skeptic verdict system was fully removed (repo PR #8217 + workflow registrations disabled + .gemini toml deletion PR #8313). This skill is retained as a historical record only — do NOT dispatch skeptic verdicts or wait for them.\n\n# Skeptic Agent — Exit Criteria Verification\n\n## When This Skill Activates\n\n**Proactively activate when:**\n- User requests a non-trivial task (multi-file, multi-step, integration work)\n- User asks to \"build\", \"implement\", \"fix\", or \"migrate\" something that will take >10 tool calls\n- Task involves E2E verification, pipeline work, or multi-system coordination\n- User invokes `\u002Fskeptic` or asks for skeptic verification\n\n**Before starting the task, ask:**\n> \"This looks non-trivial. Want to define skeptic exit criteria? A separate agent will independently verify these when you think you're done. What does 'actually done' look like for this task?\"\n\nIf the user declines, proceed without. If they define criteria, save them to `specs\u002Fexit-criteria.md` in the workspace.\n\n## How It Works\n\n### 1. Define Exit Criteria (human writes, at task start)\n\n```markdown\n# specs\u002Fexit-criteria.md\n\n## Task: [task name]\n\n### Criterion A: [name]\nWhat to verify: [natural language description]\nCommand to run (if applicable): [exact command]\nWhat PASS looks like: [expected output\u002Fstate]\nWhat FAIL looks like: [common proxy substitutions to watch for]\n\n### Criterion B: [name]\n...\n```\n\n### 2. Coder Works (normal agent session)\n\nThe coding agent works normally. It does NOT see `specs\u002Fexit-criteria.md`.\nIt signals readiness by stating \"I believe the task is complete\" or similar.\n\n### 3. Skeptic Evaluates (separate session, different model if possible)\n\nWhen the coder signals completion, spawn or switch to a skeptic session.\nThe skeptic's system instructions:\n\n```\nYou are a QA Skeptic. Your job is to FIND GAPS in the implementation.\n\nINVERTED INCENTIVE: You are rewarded for finding missing evidence.\nA false PASS is YOUR failure. A thorough FAIL report is success.\n\nRules:\n1. Read specs\u002Fexit-criteria.md\n2. For each criterion, run the EXACT verification specified\n3. Do NOT accept the coder's claims — verify independently\n4. Unit tests do NOT satisfy E2E criteria\n5. Manual tool calls do NOT satisfy pipeline criteria\n6. \"Code compiles\" does NOT mean \"feature works\"\n7. EVIDENCE MUST BE VIDEO: For UI and interactive terminal criteria, ONLY accept .gif, .mp4, or .cast videos tied to a commit SHA. Reject static screenshots unless explicitly justified.\n\nOutput format per criterion:\nCRITERION: [quote verbatim from specs\u002Fexit-criteria.md]\nEVIDENCE FOUND: [what you actually observed — commands run, output seen]\nEVIDENCE MISSING: [what should exist but doesn't]\nVERDICT: PASS | FAIL | INSUFFICIENT\nREASON: [specific gap or confirmation]\n```\n\n### 4. Feedback Loop (coder iterates on skeptic findings)\n\nIf ANY criterion is FAIL or INSUFFICIENT:\n1. Skeptic's findings are injected into the coder's next prompt\n2. Coder addresses ONLY the specific gaps identified\n3. Skeptic re-evaluates\n4. Loop until all criteria PASS or human intervenes\n\n### 5. Final Report\n\n```markdown\n## Skeptic Verification Report\n\nTask: [name]\nDate: [date]\nCoder model: [model]\nSkeptic model: [model]\nIterations: [N]\n\n| Criterion | Verdict | Evidence |\n|---|---|---|\n| A | PASS | [brief evidence] |\n| B | FAIL | [what's missing] |\n\nOverall: PASS \u002F FAIL\n```\n\n## Implementation in Different Environments\n\n### Claude Code (this environment)\n- Coder: current session\n- Skeptic: spawn via `Agent` tool with `subagent_type=\"pair-verifier\"` and skeptic system prompt\n- Or: user opens second Claude Code session with skeptic instructions\n\n### Codex CLI\n- Coder: `codex exec` with task prompt\n- Skeptic: separate `codex exec` with skeptic prompt + workspace access\n\n### AO (Agent Orchestrator)\n- Coder: `ao spawn` worker session\n- Skeptic: `ao spawn --skeptic` (new flag, spawns with skeptic agentRules)\n- Feedback loop: orchestrator mediates, injects skeptic findings into coder tmux\n\n### Ralph\u002FPair\n- Modify `\u002Fpair` verifier phase to use skeptic LLM instead of `verifyCommand` bash\n- Best of both: run `verifyCommand` first (fast, deterministic), then skeptic for nuanced criteria\n\n## Anti-Patterns\n\n- **DON'T** let the coder see the skeptic's instructions (prevents gaming)\n- **DON'T** let the coder argue with the skeptic directly (prevents rationalization)\n- **DON'T** use the same model for both (self-consistency defeats self-assessment)\n- **DON'T** skip the skeptic for \"simple\" tasks that turn out complex\n- **DON'T** let the skeptic declare PASS without running verification commands\n\n## Key Insight\n\nRLHF makes agents want to complete tasks. The Skeptic's \"task\" IS finding gaps.\nIts RLHF bias pushes it toward thoroughness in criticism, not toward premature approval.\nThis turns RLHF from a bug into a feature.\n","> **DELETED SYSTEM (2026-07-09).** The skeptic verdict system was fully removed (repo PR #8217 + workflow registrations disabled + .gemini toml deletion PR #8313). This skill is retained as a historical record only — do NOT dispatch skeptic verdicts or wait for them.\n\n# Skeptic Agent — Exit Criteria Verification\n\n## When This Skill Activates\n\n**Proactively activate when:**\n- User requests a non-trivial task (multi-file, multi-step, integration work)\n- User asks to \"build\", \"implement\", \"fix\", or \"migrate\" something that will take >10 tool calls\n- Task involves E2E verification, pipeline work, or multi-system coordination\n- User invokes `\u002Fskeptic` or asks for skeptic verification\n\n**Before starting the task, ask:**\n> \"This looks non-trivial. Want to define skeptic exit criteria? A separate agent will independently verify these when you think you're done. What does 'actually done' look like for this task?\"\n\nIf the user declines, proceed without. If they define criteria, save them to `specs\u002Fexit-criteria.md` in the workspace.\n\n## How It Works\n\n### 1. Define Exit Criteria (human writes, at task start)\n\n```markdown\n# specs\u002Fexit-criteria.md\n\n## Task: [task name]\n\n### Criterion A: [name]\nWhat to verify: [natural language description]\nCommand to run (if applicable): [exact command]\nWhat PASS looks like: [expected output\u002Fstate]\nWhat FAIL looks like: [common proxy substitutions to watch for]\n\n### Criterion B: [name]\n...\n```\n\n### 2. Coder Works (normal agent session)\n\nThe coding agent works normally. It does NOT see `specs\u002Fexit-criteria.md`.\nIt signals readiness by stating \"I believe the task is complete\" or similar.\n\n### 3. Skeptic Evaluates (separate session, different model if possible)\n\nWhen the coder signals completion, spawn or switch to a skeptic session.\nThe skeptic's system instructions:\n\n```\nYou are a QA Skeptic. Your job is to FIND GAPS in the implementation.\n\nINVERTED INCENTIVE: You are rewarded for finding missing evidence.\nA false PASS is YOUR failure. A thorough FAIL report is success.\n\nRules:\n1. Read specs\u002Fexit-criteria.md\n2. For each criterion, run the EXACT verification specified\n3. Do NOT accept the coder's claims — verify independently\n4. Unit tests do NOT satisfy E2E criteria\n5. Manual tool calls do NOT satisfy pipeline criteria\n6. \"Code compiles\" does NOT mean \"feature works\"\n7. EVIDENCE MUST BE VIDEO: For UI and interactive terminal criteria, ONLY accept .gif, .mp4, or .cast videos tied to a commit SHA. Reject static screenshots unless explicitly justified.\n\nOutput format per criterion:\nCRITERION: [quote verbatim from specs\u002Fexit-criteria.md]\nEVIDENCE FOUND: [what you actually observed — commands run, output seen]\nEVIDENCE MISSING: [what should exist but doesn't]\nVERDICT: PASS | FAIL | INSUFFICIENT\nREASON: [specific gap or confirmation]\n```\n\n### 4. Feedback Loop (coder iterates on skeptic findings)\n\nIf ANY criterion is FAIL or INSUFFICIENT:\n1. Skeptic's findings are injected into the coder's next prompt\n2. Coder addresses ONLY the specific gaps identified\n3. Skeptic re-evaluates\n4. Loop until all criteria PASS or human intervenes\n\n### 5. Final Report\n\n```markdown\n## Skeptic Verification Report\n\nTask: [name]\nDate: [date]\nCoder model: [model]\nSkeptic model: [model]\nIterations: [N]\n\n| Criterion | Verdict | Evidence |\n|---|---|---|\n| A | PASS | [brief evidence] |\n| B | FAIL | [what's missing] |\n\nOverall: PASS \u002F FAIL\n```\n\n## Implementation in Different Environments\n\n### Claude Code (this environment)\n- Coder: current session\n- Skeptic: spawn via `Agent` tool with `subagent_type=\"pair-verifier\"` and skeptic system prompt\n- Or: user opens second Claude Code session with skeptic instructions\n\n### Codex CLI\n- Coder: `codex exec` with task prompt\n- Skeptic: separate `codex exec` with skeptic prompt + workspace access\n\n### AO (Agent Orchestrator)\n- Coder: `ao spawn` worker session\n- Skeptic: `ao spawn --skeptic` (new flag, spawns with skeptic agentRules)\n- Feedback loop: orchestrator mediates, injects skeptic findings into coder tmux\n\n### Ralph\u002FPair\n- Modify `\u002Fpair` verifier phase to use skeptic LLM instead of `verifyCommand` bash\n- Best of both: run `verifyCommand` first (fast, deterministic), then skeptic for nuanced criteria\n\n## Anti-Patterns\n\n- **DON'T** let the coder see the skeptic's instructions (prevents gaming)\n- **DON'T** let the coder argue with the skeptic directly (prevents rationalization)\n- **DON'T** use the same model for both (self-consistency defeats self-assessment)\n- **DON'T** skip the skeptic for \"simple\" tasks that turn out complex\n- **DON'T** let the skeptic declare PASS without running verification commands\n\n## Key Insight\n\nRLHF makes agents want to complete tasks. The Skeptic's \"task\" IS finding gaps.\nIts RLHF bias pushes it toward thoroughness in criticism, not toward premature approval.\nThis turns RLHF from a bug into a feature.\n",{"name":5,"description":7},"main","42ccd02cbf5e8e47839dc3eab6fc754dc6b339cf",".claude\u002Fskills_archive\u002F2026-08-27-historical-zero-use\u002Fskeptic-agent",0,"https:\u002F\u002Fgithub.com\u002Fjleechanorg\u002Fjleechan-skills","https:\u002F\u002Fgithub.com\u002Fjleechanorg","2026-11-12 14:45:07",{"source":8,"skillName":5,"commands":47,"prompt":57,"zipAvailable":58},[48,51,54],{"tool":49,"command":50},"npx","npx skills add jleechanorg\u002Fjleechan-skills --skill skeptic-agent",{"tool":52,"command":53},"bunx","bunx skills add jleechanorg\u002Fjleechan-skills --skill skeptic-agent",{"tool":55,"command":56},"pnpm","pnpm dlx skills add jleechanorg\u002Fjleechan-skills --skill skeptic-agent","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fjleechanorg\u002Fjleechan-skills\u002Fblob\u002Fmain\u002F.claude\u002Fskills_archive\u002F2026-08-27-historical-zero-use\u002Fskeptic-agent\u002FSKILL.md",true,[],[61,79,100,114,128,144],{"skillNo":62,"name":63,"nameZh":6,"description":64,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":65,"categories":66,"keywords":67,"publishedAt":34},"01M716Y5MTJ6GYJZ2ATXCJK8G4","instinct","Use when the user invokes \u002Finstinct or \u002Fin, or asks to \"message Instinct\", \"ask Instinct\", \"tell Instinct\", \"check Instinct\", or read replies from the configured peer assistant. iMessage first, WhatsApp Web as fallback.","https:\u002F\u002Fgithub.com\u002Fjleechanorg\u002Fjleechan-skills\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Finstinct\u002FSKILL.md",[],[63,68,69,70,71,72,73,74,75,76,77,78],"user","when","replies","check","use","identity","whatsapp","gmail","read","message","ask",{"skillNo":80,"name":81,"nameZh":6,"description":82,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":83,"categories":84,"keywords":90,"publishedAt":34},"01M716Y4HF7FVVMJSA2CCVN5VF","hr","Compatibility alias for history-resume; produce the same fresh-state, provenance-aware resume briefing.","https:\u002F\u002Fgithub.com\u002Fjleechanorg\u002Fjleechan-skills\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fhr\u002FSKILL.md",[85],{"categoryNo":86,"categoryName":87,"taxonomyType":88,"taxonomyTypeDesc":89},"01JBSEEDCAT000000000000038","人力资源","PROFESSION","适用职业",[81,91,92,93,32,94,95,96,97,98,99],"resume","compatibility","provenance","history","produce","alias","briefing","state","fresh",{"skillNo":101,"name":102,"nameZh":6,"description":103,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":104,"categories":105,"keywords":107,"publishedAt":34},"01M716Y1FFNFKFJYYKF8HQA6T8","history-resume","Resume briefing for the current repository branch or worktree using fresh state and admitted history sources.","https:\u002F\u002Fgithub.com\u002Fjleechanorg\u002Fjleechan-skills\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fhistory-resume\u002FSKILL.md",[106],{"categoryNo":86,"categoryName":87,"taxonomyType":88,"taxonomyTypeDesc":89},[94,91,108,109,68,110,97,111,32,112,113,98,99],"history resume","branch","analysis","repository","worktree","sources",{"skillNo":115,"name":116,"nameZh":6,"description":117,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":118,"categories":119,"keywords":120,"publishedAt":34},"01M716XWXJ8EPHK4RH9T14TR85","grok","Alias for the canonical grokbot browser conversation workflow; \u002Fgrok talks to the Grok bot the same way \u002Fdot talks to Dot.","https:\u002F\u002Fgithub.com\u002Fjleechanorg\u002Fjleechan-skills\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fgrok\u002FSKILL.md",[],[116,121,122,123,32,124,96,125,126,127],"dot","bot","workflow","canonical","conversation","browser","grokbot",{"skillNo":129,"name":130,"nameZh":6,"description":131,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":132,"categories":133,"keywords":134,"publishedAt":34},"01M70N9Q07DTE0EW5KFXMMMSJ0","codexa-bench","Compare `codexa` and normal `codex` on a fair split of comparable work by default; use `--strict` for duplicate runs of the exact same task.","https:\u002F\u002Fgithub.com\u002Fjleechanorg\u002Fjleechan-skills\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fcodexa-bench\u002FSKILL.md",[],[135,136,137,72,138,32,139,140,141,142,143],"codexa bench","work","split","default","duplicate","task","compare","normal","codex",{"skillNo":145,"name":146,"nameZh":6,"description":147,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":148,"licenseFileUrl":149,"sourceHtmlUrl":150,"categories":151,"keywords":161,"publishedAt":34},"01M6C24S2X60SP8BBFGAEKPEGW","frontend-design","Create distinctive, production-grade frontend interfaces with high design quality. Use this skill when the user asks to build web components, pages, artifacts, posters, or applications (examples include websites, landing pages, dashboards, React components, HTML\u002FCSS layouts, or when styling\u002Fbeautifying any web UI). Generates creative, polished code and UI design that avoids generic AI aesthetics.","Apache-2.0","https:\u002F\u002Fgithub.com\u002Fjleechanorg\u002Fjleechan-skills\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Ffrontend-design\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002Fjleechanorg\u002Fjleechan-skills\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Ffrontend-design\u002FSKILL.md",[152,155,158],{"categoryNo":153,"categoryName":154,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000009","前端开发",{"categoryNo":156,"categoryName":157,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000015","视觉设计",{"categoryNo":159,"categoryName":160,"taxonomyType":88,"taxonomyTypeDesc":89},"01JBSEEDCAT000000000000032","设计师",[162,163,164,72,165,166,167,68,168,169,170,171,172,173,174,175,176,177,178,179,180,181,182,31,183,184,69,185],"frontend","design","frontend design","generic","ui design","react","html","components","code","artifacts","ui","aesthetics","Production","web components","landing","applications","quality","web","css","create","examples","react components","web ui","pages"]