[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6DT005YAPNDFMH28TKXECB6":3,"skill-same:01M6DT005YAPNDFMH28TKXECB6":57},{"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":17,"publishedAt":29,"linkOnly":30,"markdownContent":31,"markdownBody":32,"frontMatter":33,"refName":36,"commitSha":37,"skillDirectory":38,"forksCount":39,"repositoryHtmlUrl":40,"repositoryHomepageUrl":6,"ownerHtmlUrl":41,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":42,"install":43,"resources":56},"01M6DT005YAPNDFMH28TKXECB6","debate",null,"Structured AI debate templates and synthesis. Use when orchestrating multi-round debates between AI tools, 'debate topic', 'argue about', 'stress test idea', 'devil advocate'.","AntifragileTech\u002Fantifragile-claude-code","AntifragileTech","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F174357194?v=4","01M6DQCYMK4CDSQQAN0ZC72NTV","01M6DQCYS9YPW1WKMQ5TJMPQ43",9,"MIT","https:\u002F\u002Fgithub.com\u002FAntifragileTech\u002Fantifragile-claude-code\u002Fblob\u002Fmaster\u002Fassets\u002Fskills\u002Fthinking\u002Fdebate\u002FSKILL.md",[],[5,18,19,20,21,22,23,24,25,26,27,28],"test","ai tools","templates","Tools","devil","tool","stress test","use","Synthesis","idea","structured","2026-11-10 10:12:30",false,"---\nname: debate\ndescription: \"Structured AI debate templates and synthesis. Use when orchestrating multi-round debates between AI tools, 'debate topic', 'argue about', 'stress test idea', 'devil advocate'.\"\nversion: 5.1.0\nargument-hint: \"[topic] [--proposer=tool] [--challenger=tool] [--rounds=N] [--effort=level]\"\n---\n\n# debate\n\nPrompt templates, context assembly rules, and synthesis format for structured multi-round debates between AI tools.\n\n## Arguments\n\nParse from `$ARGUMENTS`:\n- **topic**: The debate question\u002Ftopic (required)\n- **--proposer**: Tool for the proposer role (claude, gemini, codex, opencode, copilot)\n- **--challenger**: Tool for the challenger role (must differ from proposer)\n- **--rounds**: Number of back-and-forth rounds (1-5, default: 2)\n- **--effort**: Thinking effort applied to all tool invocations (low, medium, high, max)\n- **--model-proposer**: Specific model for proposer (optional)\n- **--model-challenger**: Specific model for challenger (optional)\n\n## Universal Rules\n\nALL participants (proposer AND challenger) MUST support claims with specific evidence (file path, code pattern, benchmark, or documented behavior). Unsupported claims from either side will be flagged by the other participant and noted in the verdict. This applies to every round.\n\n## Prompt Templates\n\n### Round 1: Proposer Opening\n\n```\nYou are participating in a structured debate as the PROPOSER.\n\nTopic: {topic}\n\nYour job: Analyze this topic thoroughly and present your position. Take a clear stance. Do not hedge excessively.\n\nYou MUST support each claim with specific evidence (file path, code pattern, benchmark, or documented behavior). Unsupported claims will be challenged. \"I think\" or \"generally speaking\" without evidence is not acceptable.\n\nProvide your analysis:\n```\n\n### Round 1: Challenger Response\n\n```\nYou are participating in a structured debate as the CHALLENGER.\n\nTopic: {topic}\n\nThe PROPOSER ({proposer_tool}) argued:\n\n---\n{proposer_round1_response}\n---\n\nYour job: Find weaknesses, blind spots, and flaws in the proposer's argument. You MUST identify at least one genuine flaw or overlooked consideration before agreeing on anything. Propose concrete alternatives where you disagree.\n\nRules:\n- Do NOT say \"great point\" or validate the proposer's reasoning before critiquing it\n- Lead with what's WRONG or MISSING, then acknowledge what's right\n- If you genuinely agree on a point, explain what RISK remains despite the agreement\n- Propose at least one concrete alternative approach\n- You MUST address at least these categories: correctness, security implications, and developer experience\n- Do NOT agree with ANY claim unless you can cite specific evidence (file path, code pattern, or documented behavior) that supports the agreement. Unsupported agreement is not allowed.\n- If the proposer makes a claim without evidence, call it out: \"This claim is unsupported.\"\n\nProvide your challenge:\n```\n\n### Round 2+: Proposer Defense\n\n```\nYou are the PROPOSER in round {round} of a structured debate.\n\nTopic: {topic}\n\n{context_summary}\n\nThe CHALLENGER ({challenger_tool}) raised these points in round {previous_round}:\n\n---\n{challenger_previous_response}\n---\n\nYour job: Address each challenge directly. For each point:\n- If they're right, concede explicitly and explain how your position evolves\n- If they're wrong, explain why with specific evidence (file path, code pattern, benchmark, or documented behavior)\n- If it's a tradeoff, acknowledge the tradeoff and explain why you still favor your approach with evidence\n\nEvery claim you make -- whether concession, rebuttal, or new argument -- MUST cite specific evidence. The challenger will reject unsupported claims.\n\nDo NOT simply restate your original position. Your response must show you engaged with the specific challenges raised.\n\nProvide your defense:\n```\n\n### Round 2+: Challenger Follow-up\n\n```\nYou are the CHALLENGER in round {round} of a structured debate.\n\nTopic: {topic}\n\n{context_summary}\n\nThe PROPOSER ({proposer_tool}) responded to your challenges:\n\n---\n{proposer_previous_response}\n---\n\nIMPORTANT: Do NOT let the proposer reframe your challenges as agreements. If they say \"we actually agree\" but haven't addressed the substance, reject it. Default to suspicion, not acceptance.\n\nYour job: Evaluate the proposer's defense. For each point they addressed:\n- Did they dodge, superficially address, or respond without evidence? Call it out: \"This defense is unsupported\" or \"This dodges the original concern\"\n- Did they concede any point? Hold them to it -- they cannot walk it back later without new evidence\n- Are there NEW weaknesses in their revised position?\n- Did they adequately address your concern with specific evidence? Only then acknowledge it, and cite what convinced you\n\nYou MUST either identify at least one new weakness or unresolved concern, OR explicitly certify a previous concern as genuinely resolved with specific evidence for why you're now satisfied. \"I'm convinced because [evidence]\" is acceptable. \"I agree now\" without evidence is not.\nIf you see new problems, raise them.\n\nProvide your follow-up:\n```\n\n## Context Assembly\n\n### Rounds 1-2: Full context\n\nInclude the full text of all prior exchanges in the prompt. Context is small enough (typically under 5000 tokens total).\n\nFormat for context block:\n```\nPrevious exchanges:\n\nRound 1 - Proposer ({proposer_tool}):\n{full response}\n\nRound 1 - Challenger ({challenger_tool}):\n{full response}\n```\n\n### Round 3+: Summarized context\n\nFor rounds 3 and beyond, replace full exchange text from rounds 1 through N-2 with a summary. Only include the most recent round's responses in full.\n\nFormat:\n```\nSummary of rounds 1-{N-2}:\n{summary of key positions, agreements, and open disagreements}\n\nRound {N-1} - Proposer ({proposer_tool}):\n{full response}\n\nRound {N-1} - Challenger ({challenger_tool}):\n{full response}\n```\n\nThe orchestrator agent (opus) generates the summary. Target: 500-800 tokens. MUST preserve:\n- Each side's core position\n- All concessions (verbatim quotes, not paraphrased)\n- All evidence citations that support agreements\n- Points of disagreement (unresolved)\n- Any contradictions between rounds (e.g., proposer concedes in round 1 but walks it back in round 2 -- note both explicitly)\n\n## Synthesis Format\n\nAfter all rounds complete, the orchestrator produces this structured output:\n\n```\n## Debate Summary\n\n**Topic**: {topic}\n**Proposer**: {proposer_tool} ({proposer_model})\n**Challenger**: {challenger_tool} ({challenger_model})\n**Rounds**: {rounds_completed}\n**Rigor**: Structured perspective comparison (prompt-enforced adversarial rules, no deterministic verification)\n\n### Verdict\n\n{winner_tool} had the stronger argument because: {specific reasoning citing debate evidence}\n\n### Debate Quality\n\nRate the debate on these dimensions:\n- **Genuine disagreement**: Did the challenger maintain independent positions, or converge toward the proposer? (high\u002Fmedium\u002Flow)\n- **Evidence quality**: Did both sides cite specific examples, or argue from generalities? (high\u002Fmedium\u002Flow)\n- **Challenge depth**: Were the challenges substantive, or surface-level? (high\u002Fmedium\u002Flow)\n\n### Key Agreements\n- {agreed point 1} (evidence: {what supports this agreement})\n- {agreed point 2} (evidence: {what supports this agreement})\n\n### Key Disagreements\n- {point}: {proposer_tool} argues {X}, {challenger_tool} argues {Y}\n\n### Unresolved Questions\n- {question that neither side adequately addressed}\n\n### Recommendation\n{Orchestrator's recommendation - must pick a direction, not \"both have merit\"}\n```\n\n**Synthesis rules:**\n- The verdict MUST pick a side. \"Both approaches have merit\" is NOT acceptable.\n- Cite specific arguments from the debate as evidence for the verdict.\n- The recommendation must be actionable - what should the user DO based on this debate.\n- Unresolved questions highlight where the debate fell short, not where both sides are \"equally valid.\"\n\n## State File Schema\n\nSave to `{AI_STATE_DIR}\u002Fdebate\u002Flast-debate.json`:\n\n```json\n{\n  \"id\": \"debate-{ISO timestamp}-{4 char random hex}\",\n  \"topic\": \"original topic text\",\n  \"proposer\": {\"tool\": \"claude\", \"model\": \"opus\"},\n  \"challenger\": {\"tool\": \"gemini\", \"model\": \"gemini-3.1-pro-preview\"},\n  \"effort\": \"high\",\n  \"rounds_completed\": 2,\n  \"max_rounds\": 2,\n  \"status\": \"completed\",\n  \"exchanges\": [\n    {\"round\": 1, \"role\": \"proposer\", \"tool\": \"claude\", \"response\": \"...\", \"duration_ms\": 8500},\n    {\"round\": 1, \"role\": \"challenger\", \"tool\": \"gemini\", \"response\": \"...\", \"duration_ms\": 12000},\n    {\"round\": 2, \"role\": \"proposer\", \"tool\": \"claude\", \"response\": \"...\", \"duration_ms\": 9200},\n    {\"round\": 2, \"role\": \"challenger\", \"tool\": \"gemini\", \"response\": \"...\", \"duration_ms\": 11000}\n  ],\n  \"verdict\": {\n    \"winner\": \"claude\",\n    \"reasoning\": \"...\",\n    \"agreements\": [\"...\"],\n    \"disagreements\": [\"...\"],\n    \"recommendation\": \"...\"\n  },\n  \"timestamp\": \"{ISO 8601 timestamp}\"\n}\n```\n\nPlatform state directory:\n- Claude Code: `.claude\u002F`\n- OpenCode: `.opencode\u002F`\n- Codex CLI: `.codex\u002F`\n\n## Error Handling\n\n| Error | Action |\n|-------|--------|\n| Proposer fails round 1 | Abort debate. Cannot proceed without opening position. |\n| Challenger fails round 1 | Show proposer's position with note: \"[WARN] Challenger failed. Showing proposer's uncontested position.\" |\n| Any tool fails mid-debate | Synthesize from completed rounds. Note incomplete round in output. |\n| Tool invocation timeout (>240s) | Round 1 proposer: abort. Round 1 challenger: proceed with uncontested. Round 2+: synthesize from completed rounds with timeout note. |\n| Consult result envelope indicates failure (status\u002Fexit\u002Ferror\u002Fempty output) | Treat as tool failure for that role\u002Fround and apply the same role+round policy above. |\n| Structured parse fails after successful envelope | Treat as tool failure for that role\u002Fround, include only sanitized parse metadata (`PARSE_ERROR:\u003Ctype>:\u003Ccode>`, redact secrets, strip control chars, max 200 chars), then apply the same role+round policy above. |\n| All rounds timeout | \"[ERROR] Debate failed: all tool invocations timed out.\" |\n| No successful exchanges recorded (non-timeout) | \"[ERROR] Debate failed: no successful exchanges were recorded.\" |\n\n## External Tool Quick Reference\n\n> Canonical source: `plugins\u002Fconsult\u002Fskills\u002Fconsult\u002FSKILL.md`. Build and execute CLI commands directly using these templates. Do NOT invoke via `Skill: consult` - in Claude Code that loads the interactive command wrapper and causes a recursive loop. Write the question to `{AI_STATE_DIR}\u002Fconsult\u002Fquestion.tmp` first, then execute the command via Bash.\n\n### Safe Command Patterns\n\n| Provider | Safe Command Pattern |\n|----------|---------------------|\n| Claude | `claude -p - --output-format json --model \"MODEL\" --max-turns TURNS --allowedTools \"Read,Glob,Grep\" \u003C \"{AI_STATE_DIR}\u002Fconsult\u002Fquestion.tmp\"` |\n| Gemini | `gemini -p - --output-format json -m \"MODEL\" \u003C \"{AI_STATE_DIR}\u002Fconsult\u002Fquestion.tmp\"` |\n| Codex | `codex exec \"$(cat \"{AI_STATE_DIR}\u002Fconsult\u002Fquestion.tmp\")\" --json -m \"MODEL\" -c model_reasoning_effort=\"LEVEL\"` |\n| OpenCode | `opencode run - --format json --model \"MODEL\" --variant \"VARIANT\" \u003C \"{AI_STATE_DIR}\u002Fconsult\u002Fquestion.tmp\"` |\n| Copilot | `copilot -p - \u003C \"{AI_STATE_DIR}\u002Fconsult\u002Fquestion.tmp\"` |\n\n### Effort-to-Model Mapping\n\n| Effort | Claude | Gemini | Codex | OpenCode | Copilot |\n|--------|--------|--------|-------|----------|---------|\n| low | claude-haiku-4-5 (1 turn) | gemini-3-flash-preview | gpt-5.3-codex (low) | default (low) | no control |\n| medium | claude-sonnet-4-6 (3 turns) | gemini-3-flash-preview | gpt-5.3-codex (medium) | default (medium) | no control |\n| high | claude-opus-4-6 (5 turns) | gemini-3.1-pro-preview | gpt-5.3-codex (high) | default (high) | no control |\n| max | claude-opus-4-6 (10 turns) | gemini-3.1-pro-preview | gpt-5.3-codex (high) | default + --thinking | no control |\n\n### Output Parsing\n\n| Provider | Parse Expression |\n|----------|-----------------|\n| Claude | `JSON.parse(stdout).result` |\n| Gemini | `JSON.parse(stdout).response` |\n| Codex | `JSON.parse(stdout).message` or raw text |\n| OpenCode | Newline-delimited JSON. Concatenate `part.text` from events where `type === \"text\"`. Session ID from `event.sessionID`. |\n| Copilot | Raw stdout text |\n\nParse discipline:\n1. Evaluate execution status first (timeout\u002Fnon-zero\u002Ferror\u002Fempty output) before any parsing.\n2. Parse only when execution status is successful.\n3. If parse fails, surface only sanitized parse metadata (never raw stdout\u002Fstderr snippets) and apply role\u002Fround failure policy instead of hanging or continuing silently.\n\n### ACP Transport Commands\n\n> ACP is an alternative transport available when providers support it. Build and execute CLI commands directly - do NOT use `Skill: consult` (recursive loop in Claude Code).\n\n| Provider | ACP Command Pattern |\n|----------|-------------------|\n| Claude | `node acp\u002Frun.js --provider=\"claude\" --question-file=\"{AI_STATE_DIR}\u002Fconsult\u002Fquestion.tmp\" --timeout=240000 --model=\"MODEL\"` |\n| Gemini | `node acp\u002Frun.js --provider=\"gemini\" --question-file=\"{AI_STATE_DIR}\u002Fconsult\u002Fquestion.tmp\" --timeout=240000 --model=\"MODEL\"` |\n| Codex | `node acp\u002Frun.js --provider=\"codex\" --question-file=\"{AI_STATE_DIR}\u002Fconsult\u002Fquestion.tmp\" --timeout=240000 --model=\"MODEL\"` |\n| OpenCode | `node acp\u002Frun.js --provider=\"opencode\" --question-file=\"{AI_STATE_DIR}\u002Fconsult\u002Fquestion.tmp\" --timeout=240000 --model=\"MODEL\"` |\n| Copilot | `node acp\u002Frun.js --provider=\"copilot\" --question-file=\"{AI_STATE_DIR}\u002Fconsult\u002Fquestion.tmp\" --timeout=240000` |\n| Kiro | `node acp\u002Frun.js --provider=\"kiro\" --question-file=\"{AI_STATE_DIR}\u002Fconsult\u002Fquestion.tmp\" --timeout=240000` |\n\nNote the 240000ms timeout (240s) for debate rounds vs 120000ms (120s) for consult.\n\n**Kiro**: ACP-only provider. No CLI mode. Available when `kiro-cli` is on PATH.\n\n### ACP Output Parsing\n\nACP transport output is parsed identically to CLI transport - the ACP runner (`acp\u002Frun.js`) normalizes responses into the same JSON envelope format. The `transport` field in the envelope indicates `\"acp\"` or `\"cli\"`.\n","# debate\n\nPrompt templates, context assembly rules, and synthesis format for structured multi-round debates between AI tools.\n\n## Arguments\n\nParse from `$ARGUMENTS`:\n- **topic**: The debate question\u002Ftopic (required)\n- **--proposer**: Tool for the proposer role (claude, gemini, codex, opencode, copilot)\n- **--challenger**: Tool for the challenger role (must differ from proposer)\n- **--rounds**: Number of back-and-forth rounds (1-5, default: 2)\n- **--effort**: Thinking effort applied to all tool invocations (low, medium, high, max)\n- **--model-proposer**: Specific model for proposer (optional)\n- **--model-challenger**: Specific model for challenger (optional)\n\n## Universal Rules\n\nALL participants (proposer AND challenger) MUST support claims with specific evidence (file path, code pattern, benchmark, or documented behavior). Unsupported claims from either side will be flagged by the other participant and noted in the verdict. This applies to every round.\n\n## Prompt Templates\n\n### Round 1: Proposer Opening\n\n```\nYou are participating in a structured debate as the PROPOSER.\n\nTopic: {topic}\n\nYour job: Analyze this topic thoroughly and present your position. Take a clear stance. Do not hedge excessively.\n\nYou MUST support each claim with specific evidence (file path, code pattern, benchmark, or documented behavior). Unsupported claims will be challenged. \"I think\" or \"generally speaking\" without evidence is not acceptable.\n\nProvide your analysis:\n```\n\n### Round 1: Challenger Response\n\n```\nYou are participating in a structured debate as the CHALLENGER.\n\nTopic: {topic}\n\nThe PROPOSER ({proposer_tool}) argued:\n\n---\n{proposer_round1_response}\n---\n\nYour job: Find weaknesses, blind spots, and flaws in the proposer's argument. You MUST identify at least one genuine flaw or overlooked consideration before agreeing on anything. Propose concrete alternatives where you disagree.\n\nRules:\n- Do NOT say \"great point\" or validate the proposer's reasoning before critiquing it\n- Lead with what's WRONG or MISSING, then acknowledge what's right\n- If you genuinely agree on a point, explain what RISK remains despite the agreement\n- Propose at least one concrete alternative approach\n- You MUST address at least these categories: correctness, security implications, and developer experience\n- Do NOT agree with ANY claim unless you can cite specific evidence (file path, code pattern, or documented behavior) that supports the agreement. Unsupported agreement is not allowed.\n- If the proposer makes a claim without evidence, call it out: \"This claim is unsupported.\"\n\nProvide your challenge:\n```\n\n### Round 2+: Proposer Defense\n\n```\nYou are the PROPOSER in round {round} of a structured debate.\n\nTopic: {topic}\n\n{context_summary}\n\nThe CHALLENGER ({challenger_tool}) raised these points in round {previous_round}:\n\n---\n{challenger_previous_response}\n---\n\nYour job: Address each challenge directly. For each point:\n- If they're right, concede explicitly and explain how your position evolves\n- If they're wrong, explain why with specific evidence (file path, code pattern, benchmark, or documented behavior)\n- If it's a tradeoff, acknowledge the tradeoff and explain why you still favor your approach with evidence\n\nEvery claim you make -- whether concession, rebuttal, or new argument -- MUST cite specific evidence. The challenger will reject unsupported claims.\n\nDo NOT simply restate your original position. Your response must show you engaged with the specific challenges raised.\n\nProvide your defense:\n```\n\n### Round 2+: Challenger Follow-up\n\n```\nYou are the CHALLENGER in round {round} of a structured debate.\n\nTopic: {topic}\n\n{context_summary}\n\nThe PROPOSER ({proposer_tool}) responded to your challenges:\n\n---\n{proposer_previous_response}\n---\n\nIMPORTANT: Do NOT let the proposer reframe your challenges as agreements. If they say \"we actually agree\" but haven't addressed the substance, reject it. Default to suspicion, not acceptance.\n\nYour job: Evaluate the proposer's defense. For each point they addressed:\n- Did they dodge, superficially address, or respond without evidence? Call it out: \"This defense is unsupported\" or \"This dodges the original concern\"\n- Did they concede any point? Hold them to it -- they cannot walk it back later without new evidence\n- Are there NEW weaknesses in their revised position?\n- Did they adequately address your concern with specific evidence? Only then acknowledge it, and cite what convinced you\n\nYou MUST either identify at least one new weakness or unresolved concern, OR explicitly certify a previous concern as genuinely resolved with specific evidence for why you're now satisfied. \"I'm convinced because [evidence]\" is acceptable. \"I agree now\" without evidence is not.\nIf you see new problems, raise them.\n\nProvide your follow-up:\n```\n\n## Context Assembly\n\n### Rounds 1-2: Full context\n\nInclude the full text of all prior exchanges in the prompt. Context is small enough (typically under 5000 tokens total).\n\nFormat for context block:\n```\nPrevious exchanges:\n\nRound 1 - Proposer ({proposer_tool}):\n{full response}\n\nRound 1 - Challenger ({challenger_tool}):\n{full response}\n```\n\n### Round 3+: Summarized context\n\nFor rounds 3 and beyond, replace full exchange text from rounds 1 through N-2 with a summary. Only include the most recent round's responses in full.\n\nFormat:\n```\nSummary of rounds 1-{N-2}:\n{summary of key positions, agreements, and open disagreements}\n\nRound {N-1} - Proposer ({proposer_tool}):\n{full response}\n\nRound {N-1} - Challenger ({challenger_tool}):\n{full response}\n```\n\nThe orchestrator agent (opus) generates the summary. Target: 500-800 tokens. MUST preserve:\n- Each side's core position\n- All concessions (verbatim quotes, not paraphrased)\n- All evidence citations that support agreements\n- Points of disagreement (unresolved)\n- Any contradictions between rounds (e.g., proposer concedes in round 1 but walks it back in round 2 -- note both explicitly)\n\n## Synthesis Format\n\nAfter all rounds complete, the orchestrator produces this structured output:\n\n```\n## Debate Summary\n\n**Topic**: {topic}\n**Proposer**: {proposer_tool} ({proposer_model})\n**Challenger**: {challenger_tool} ({challenger_model})\n**Rounds**: {rounds_completed}\n**Rigor**: Structured perspective comparison (prompt-enforced adversarial rules, no deterministic verification)\n\n### Verdict\n\n{winner_tool} had the stronger argument because: {specific reasoning citing debate evidence}\n\n### Debate Quality\n\nRate the debate on these dimensions:\n- **Genuine disagreement**: Did the challenger maintain independent positions, or converge toward the proposer? (high\u002Fmedium\u002Flow)\n- **Evidence quality**: Did both sides cite specific examples, or argue from generalities? (high\u002Fmedium\u002Flow)\n- **Challenge depth**: Were the challenges substantive, or surface-level? (high\u002Fmedium\u002Flow)\n\n### Key Agreements\n- {agreed point 1} (evidence: {what supports this agreement})\n- {agreed point 2} (evidence: {what supports this agreement})\n\n### Key Disagreements\n- {point}: {proposer_tool} argues {X}, {challenger_tool} argues {Y}\n\n### Unresolved Questions\n- {question that neither side adequately addressed}\n\n### Recommendation\n{Orchestrator's recommendation - must pick a direction, not \"both have merit\"}\n```\n\n**Synthesis rules:**\n- The verdict MUST pick a side. \"Both approaches have merit\" is NOT acceptable.\n- Cite specific arguments from the debate as evidence for the verdict.\n- The recommendation must be actionable - what should the user DO based on this debate.\n- Unresolved questions highlight where the debate fell short, not where both sides are \"equally valid.\"\n\n## State File Schema\n\nSave to `{AI_STATE_DIR}\u002Fdebate\u002Flast-debate.json`:\n\n```json\n{\n  \"id\": \"debate-{ISO timestamp}-{4 char random hex}\",\n  \"topic\": \"original topic text\",\n  \"proposer\": {\"tool\": \"claude\", \"model\": \"opus\"},\n  \"challenger\": {\"tool\": \"gemini\", \"model\": \"gemini-3.1-pro-preview\"},\n  \"effort\": \"high\",\n  \"rounds_completed\": 2,\n  \"max_rounds\": 2,\n  \"status\": \"completed\",\n  \"exchanges\": [\n    {\"round\": 1, \"role\": \"proposer\", \"tool\": \"claude\", \"response\": \"...\", \"duration_ms\": 8500},\n    {\"round\": 1, \"role\": \"challenger\", \"tool\": \"gemini\", \"response\": \"...\", \"duration_ms\": 12000},\n    {\"round\": 2, \"role\": \"proposer\", \"tool\": \"claude\", \"response\": \"...\", \"duration_ms\": 9200},\n    {\"round\": 2, \"role\": \"challenger\", \"tool\": \"gemini\", \"response\": \"...\", \"duration_ms\": 11000}\n  ],\n  \"verdict\": {\n    \"winner\": \"claude\",\n    \"reasoning\": \"...\",\n    \"agreements\": [\"...\"],\n    \"disagreements\": [\"...\"],\n    \"recommendation\": \"...\"\n  },\n  \"timestamp\": \"{ISO 8601 timestamp}\"\n}\n```\n\nPlatform state directory:\n- Claude Code: `.claude\u002F`\n- OpenCode: `.opencode\u002F`\n- Codex CLI: `.codex\u002F`\n\n## Error Handling\n\n| Error | Action |\n|-------|--------|\n| Proposer fails round 1 | Abort debate. Cannot proceed without opening position. |\n| Challenger fails round 1 | Show proposer's position with note: \"[WARN] Challenger failed. Showing proposer's uncontested position.\" |\n| Any tool fails mid-debate | Synthesize from completed rounds. Note incomplete round in output. |\n| Tool invocation timeout (>240s) | Round 1 proposer: abort. Round 1 challenger: proceed with uncontested. Round 2+: synthesize from completed rounds with timeout note. |\n| Consult result envelope indicates failure (status\u002Fexit\u002Ferror\u002Fempty output) | Treat as tool failure for that role\u002Fround and apply the same role+round policy above. |\n| Structured parse fails after successful envelope | Treat as tool failure for that role\u002Fround, include only sanitized parse metadata (`PARSE_ERROR:\u003Ctype>:\u003Ccode>`, redact secrets, strip control chars, max 200 chars), then apply the same role+round policy above. |\n| All rounds timeout | \"[ERROR] Debate failed: all tool invocations timed out.\" |\n| No successful exchanges recorded (non-timeout) | \"[ERROR] Debate failed: no successful exchanges were recorded.\" |\n\n## External Tool Quick Reference\n\n> Canonical source: `plugins\u002Fconsult\u002Fskills\u002Fconsult\u002FSKILL.md`. Build and execute CLI commands directly using these templates. Do NOT invoke via `Skill: consult` - in Claude Code that loads the interactive command wrapper and causes a recursive loop. Write the question to `{AI_STATE_DIR}\u002Fconsult\u002Fquestion.tmp` first, then execute the command via Bash.\n\n### Safe Command Patterns\n\n| Provider | Safe Command Pattern |\n|----------|---------------------|\n| Claude | `claude -p - --output-format json --model \"MODEL\" --max-turns TURNS --allowedTools \"Read,Glob,Grep\" \u003C \"{AI_STATE_DIR}\u002Fconsult\u002Fquestion.tmp\"` |\n| Gemini | `gemini -p - --output-format json -m \"MODEL\" \u003C \"{AI_STATE_DIR}\u002Fconsult\u002Fquestion.tmp\"` |\n| Codex | `codex exec \"$(cat \"{AI_STATE_DIR}\u002Fconsult\u002Fquestion.tmp\")\" --json -m \"MODEL\" -c model_reasoning_effort=\"LEVEL\"` |\n| OpenCode | `opencode run - --format json --model \"MODEL\" --variant \"VARIANT\" \u003C \"{AI_STATE_DIR}\u002Fconsult\u002Fquestion.tmp\"` |\n| Copilot | `copilot -p - \u003C \"{AI_STATE_DIR}\u002Fconsult\u002Fquestion.tmp\"` |\n\n### Effort-to-Model Mapping\n\n| Effort | Claude | Gemini | Codex | OpenCode | Copilot |\n|--------|--------|--------|-------|----------|---------|\n| low | claude-haiku-4-5 (1 turn) | gemini-3-flash-preview | gpt-5.3-codex (low) | default (low) | no control |\n| medium | claude-sonnet-4-6 (3 turns) | gemini-3-flash-preview | gpt-5.3-codex (medium) | default (medium) | no control |\n| high | claude-opus-4-6 (5 turns) | gemini-3.1-pro-preview | gpt-5.3-codex (high) | default (high) | no control |\n| max | claude-opus-4-6 (10 turns) | gemini-3.1-pro-preview | gpt-5.3-codex (high) | default + --thinking | no control |\n\n### Output Parsing\n\n| Provider | Parse Expression |\n|----------|-----------------|\n| Claude | `JSON.parse(stdout).result` |\n| Gemini | `JSON.parse(stdout).response` |\n| Codex | `JSON.parse(stdout).message` or raw text |\n| OpenCode | Newline-delimited JSON. Concatenate `part.text` from events where `type === \"text\"`. Session ID from `event.sessionID`. |\n| Copilot | Raw stdout text |\n\nParse discipline:\n1. Evaluate execution status first (timeout\u002Fnon-zero\u002Ferror\u002Fempty output) before any parsing.\n2. Parse only when execution status is successful.\n3. If parse fails, surface only sanitized parse metadata (never raw stdout\u002Fstderr snippets) and apply role\u002Fround failure policy instead of hanging or continuing silently.\n\n### ACP Transport Commands\n\n> ACP is an alternative transport available when providers support it. Build and execute CLI commands directly - do NOT use `Skill: consult` (recursive loop in Claude Code).\n\n| Provider | ACP Command Pattern |\n|----------|-------------------|\n| Claude | `node acp\u002Frun.js --provider=\"claude\" --question-file=\"{AI_STATE_DIR}\u002Fconsult\u002Fquestion.tmp\" --timeout=240000 --model=\"MODEL\"` |\n| Gemini | `node acp\u002Frun.js --provider=\"gemini\" --question-file=\"{AI_STATE_DIR}\u002Fconsult\u002Fquestion.tmp\" --timeout=240000 --model=\"MODEL\"` |\n| Codex | `node acp\u002Frun.js --provider=\"codex\" --question-file=\"{AI_STATE_DIR}\u002Fconsult\u002Fquestion.tmp\" --timeout=240000 --model=\"MODEL\"` |\n| OpenCode | `node acp\u002Frun.js --provider=\"opencode\" --question-file=\"{AI_STATE_DIR}\u002Fconsult\u002Fquestion.tmp\" --timeout=240000 --model=\"MODEL\"` |\n| Copilot | `node acp\u002Frun.js --provider=\"copilot\" --question-file=\"{AI_STATE_DIR}\u002Fconsult\u002Fquestion.tmp\" --timeout=240000` |\n| Kiro | `node acp\u002Frun.js --provider=\"kiro\" --question-file=\"{AI_STATE_DIR}\u002Fconsult\u002Fquestion.tmp\" --timeout=240000` |\n\nNote the 240000ms timeout (240s) for debate rounds vs 120000ms (120s) for consult.\n\n**Kiro**: ACP-only provider. No CLI mode. Available when `kiro-cli` is on PATH.\n\n### ACP Output Parsing\n\nACP transport output is parsed identically to CLI transport - the ACP runner (`acp\u002Frun.js`) normalizes responses into the same JSON envelope format. The `transport` field in the envelope indicates `\"acp\"` or `\"cli\"`.\n",{"name":5,"version":34,"description":7,"argument-hint":35},"5.1.0","[topic] [--proposer=tool] [--challenger=tool] [--rounds=N] [--effort=level]","master","c73386b305dec167db7654e0e6feacb462d4f9dc","assets\u002Fskills\u002Fthinking\u002Fdebate",1,"https:\u002F\u002Fgithub.com\u002FAntifragileTech\u002Fantifragile-claude-code","https:\u002F\u002Fgithub.com\u002FAntifragileTech","2026-10-02 14:22:14",{"source":8,"skillName":5,"commands":44,"prompt":54,"zipAvailable":55},[45,48,51],{"tool":46,"command":47},"npx","npx skills add AntifragileTech\u002Fantifragile-claude-code --skill debate",{"tool":49,"command":50},"bunx","bunx skills add AntifragileTech\u002Fantifragile-claude-code --skill debate",{"tool":52,"command":53},"pnpm","pnpm dlx skills add AntifragileTech\u002Fantifragile-claude-code --skill debate","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FAntifragileTech\u002Fantifragile-claude-code\u002Fblob\u002Fmaster\u002Fassets\u002Fskills\u002Fthinking\u002Fdebate\u002FSKILL.md",true,[],[58,83,98,113,126,139],{"skillNo":59,"name":60,"nameZh":6,"description":61,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":62,"categories":63,"keywords":77,"publishedAt":29},"01M6DT9Q6VYA615SCXR757TZZ7","writing-skills","Use when creating new skills, editing existing skills, or verifying skills work before deployment","https:\u002F\u002Fgithub.com\u002FAntifragileTech\u002Fantifragile-claude-code\u002Fblob\u002Fmaster\u002Fassets\u002Fskills\u002Fthinking\u002Fwriting-skills\u002FSKILL.md",[64,69,72],{"categoryNo":65,"categoryName":66,"taxonomyType":67,"taxonomyTypeDesc":68},"01JBSEEDCAT000000000000008","测试与质量","SCENARIO","使用场景",{"categoryNo":70,"categoryName":71,"taxonomyType":67,"taxonomyTypeDesc":68},"01JBSEEDCAT000000000000018","写作与文案",{"categoryNo":73,"categoryName":74,"taxonomyType":75,"taxonomyTypeDesc":76},"01JBSEEDCAT000000000000036","内容创作者","PROFESSION","适用职业",[78,79,25,80,81,82],"writing skills","writing","editing","work","deployment",{"skillNo":84,"name":85,"nameZh":6,"description":86,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":87,"categories":88,"keywords":91,"publishedAt":29},"01M6DT9GFHDNY90P41WGAE1TV0","writing-plans","Use when you have a spec or requirements for a multi-step task, before touching code","https:\u002F\u002Fgithub.com\u002FAntifragileTech\u002Fantifragile-claude-code\u002Fblob\u002Fmaster\u002Fassets\u002Fskills\u002Fthinking\u002Fwriting-plans\u002FSKILL.md",[89,90],{"categoryNo":70,"categoryName":71,"taxonomyType":67,"taxonomyTypeDesc":68},{"categoryNo":73,"categoryName":74,"taxonomyType":75,"taxonomyTypeDesc":76},[92,93,79,94,95,25,96,97],"plans","writing plans","requirements","task","code","spec",{"skillNo":99,"name":100,"nameZh":6,"description":101,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":102,"categories":103,"keywords":106,"publishedAt":29},"01M6DT9C6PFGD86WS3HY7YE3XG","docs:write-concisely","Apply writing rules to any documentation that humans will read. Makes your writing clearer, stronger, and more professional.","https:\u002F\u002Fgithub.com\u002FAntifragileTech\u002Fantifragile-claude-code\u002Fblob\u002Fmaster\u002Fassets\u002Fskills\u002Fthinking\u002Fwrite-concisely\u002FSKILL.md",[104,105],{"categoryNo":70,"categoryName":71,"taxonomyType":67,"taxonomyTypeDesc":68},{"categoryNo":73,"categoryName":74,"taxonomyType":75,"taxonomyTypeDesc":76},[107,108,109,110,111,112,79],"docs:write concisely","write","docs","read","Rules","documentation",{"skillNo":114,"name":115,"nameZh":6,"description":116,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":117,"categories":118,"keywords":119,"publishedAt":29},"01M6DT98AWQ27P0N1CG97WAG66","kaizen:why","Iterative Five Whys root cause analysis drilling from symptoms to fundamentals","https:\u002F\u002Fgithub.com\u002FAntifragileTech\u002Fantifragile-claude-code\u002Fblob\u002Fmaster\u002Fassets\u002Fskills\u002Fthinking\u002Fwhy\u002FSKILL.md",[],[115,120,121,122,123,124,125],"why","kaizen","issue","five whys","analysis","root cause analysis",{"skillNo":127,"name":128,"nameZh":6,"description":129,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":130,"categories":131,"keywords":132,"publishedAt":29},"01M6DT93EH25NAX19R3WGE1P53","verification-loop","A comprehensive verification system for Claude Code sessions.","https:\u002F\u002Fgithub.com\u002FAntifragileTech\u002Fantifragile-claude-code\u002Fblob\u002Fmaster\u002Fassets\u002Fskills\u002Fthinking\u002Fverification-loop\u002FSKILL.md",[],[133,134,135,136,96,137,138],"verification loop","verification","loop","sessions","claude code sessions","system",{"skillNo":140,"name":141,"nameZh":6,"description":142,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":143,"categories":144,"keywords":145,"publishedAt":29},"01M6DT90GW7Z7VCN98P9DVYSNC","verification-before-completion","Use when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always","https:\u002F\u002Fgithub.com\u002FAntifragileTech\u002Fantifragile-claude-code\u002Fblob\u002Fmaster\u002Fassets\u002Fskills\u002Fthinking\u002Fverification-before-completion\u002FSKILL.md",[],[146,134,147,81,25,148,149,150],"verification before completion","commands","evidence","claim","complete"]