[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M73DWZCM7017M9RPCTZAVGK2":3,"skill-same:01M73DWZCM7017M9RPCTZAVGK2":47},{"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":21,"frontMatter":22,"refName":23,"commitSha":24,"skillDirectory":25,"forksCount":26,"repositoryHtmlUrl":27,"repositoryHomepageUrl":28,"ownerHtmlUrl":29,"ownerBlogUrl":30,"ownerDescription":31,"updatedAt":32,"install":33,"resources":46},"01M73DWZCM7017M9RPCTZAVGK2","multi-agent-debate",null,"Horonomos-laps\u002FYana-AI","Horonomos-laps","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F355299122?v=4","01M76B3QPXE88ET2SQPDPVATZT"," Horonomos-laps","01M72ZQRJPCYPSSWT4D00VJR2A",13,"Apache-2.0","https:\u002F\u002Fgithub.com\u002FHoronomos-laps\u002FYana-AI\u002Fblob\u002Fmain\u002Fcore\u002Fskills\u002Fmulti-agent-debate\u002FSKILL.md",[],[],"2026-11-11 15:11:14",false,"---\nname: multi-agent-debate\ndescription: Multi-agent debate \u002F Society of Mind — N agents × R rounds converge on consensus. Du et al. 2023 beats zero-shot CoT on MMLU\u002FGSM8K\u002FMATH. Independent contributions from agent count AND round count. Sycophancy cascade prevention, heterogeneous models, compute budget. Sources: rohitg00\u002Fai-engineering-from-scratch (Apache-2.0).\norigin: yana-ai — synthesized from rohitg00\u002Fai-engineering-from-scratch (Apache-2.0)\nlicense: Apache-2.0\nversion: 1.0.0\ncompatibility: yana-ai >= 1.3.54\n---\n\n# \u002Fmulti-agent-debate\n\n## When to Use\n\n- High-stakes reasoning where a single LLM gives confidently wrong answers\n- Reducing correlated errors: one model family has systematic bias in a domain\n- Factual tasks where self-consistency sampling has plateaued (debate breaks the saturation)\n- Research synthesis, code review, or multi-perspective analysis\n\n## Do NOT use for\n\n- Tasks with a single deterministic answer verifiable by tool (use CRITIC in [[self-refine-critic]])\n- Latency-sensitive workloads (N × R LLM calls = N×R cost)\n- Tasks where any model in the pool is clearly much weaker than others (weak models drag consensus)\n\n---\n\n## Du et al. 2023 algorithm (ICML 2024)\n\n```\nRound 0 (initial):  each agent i produces answer_i independently\nRound r (debate):   each agent i sees all agents' round r-1 answers\n                    → produces updated answer_i conditioned on the group\nFinal:              majority-vote across agents' final answers\n```\n\nTwo independent contributions to quality (from paper ablations):\n- **Multiple agents** (1 round, majority vote): beats single-agent; plateaus\n- **Multiple rounds** (1 agent, self-reflection): barely helps alone\n- **Both together**: produces the big accuracy jumps\n\n---\n\n## Debate loop\n\n```python\nfrom typing import Callable\n\ndef debate(\n    question: str,\n    agents: list[Callable[[str, list[str] | None], str]],\n    rounds: int = 3,\n) -> str:\n    \"\"\"\n    agents: list of callables (question, prior_answers) -> answer\n    Returns majority-voted final answer.\n    \"\"\"\n    # Round 0: independent proposals\n    answers: list[str] = [agent(question, None) for agent in agents]\n\n    # Rounds 1..R: each agent sees all others' prior answers\n    for r in range(1, rounds):\n        new_answers: list[str] = []\n        for i, agent in enumerate(agents):\n            others = [answers[j] for j in range(len(agents)) if j != i]\n            updated = agent(question, others)\n            new_answers.append(updated)\n        answers = new_answers\n\n    return majority_vote(answers)\n\ndef majority_vote(answers: list[str]) -> str:\n    from collections import Counter\n    # Normalize: extract final \"Answer: X\" from each response\n    import re\n    extracted = []\n    for a in answers:\n        m = re.search(r'(?:answer|final answer)[:\\s]+(.+?)(?:\\.|$)', a, re.IGNORECASE)\n        extracted.append(m.group(1).strip() if m else a.strip())\n    counts = Counter(extracted)\n    return counts.most_common(1)[0][0]\n```\n\n---\n\n## Agent prompt with debate context\n\n```python\ndef build_debate_prompt(question: str, prior_answers: list[str] | None) -> str:\n    if not prior_answers:\n        return f\"\"\"Answer the following question. Show your reasoning.\n\nQuestion: {question}\n\nAnswer:\"\"\"\n\n    others_block = \"\\n\\n\".join(\n        f\"Agent {i+1}'s answer:\\n{ans}\"\n        for i, ans in enumerate(prior_answers)\n    )\n    return f\"\"\"Answer the following question. You have seen other agents' answers below.\nConsider their reasoning critically. You may update your answer or maintain it with justification.\n\nQuestion: {question}\n\nOther agents' answers:\n{others_block}\n\nYour updated answer (with reasoning):\"\"\"\n```\n\n---\n\n## Sycophancy cascade prevention\n\n```python\ndef build_adversarial_debate_prompt(question: str, prior_answers: list[str]) -> str:\n    \"\"\"Force at least one agent to argue the counter-position.\"\"\"\n    others_block = \"\\n\\n\".join(f\"Agent {i+1}: {a}\" for i, a in enumerate(prior_answers))\n    return f\"\"\"Question: {question}\n\nOther agents have proposed the following answers:\n{others_block}\n\nYour role: identify the weakest point in the most popular answer and argue for an alternative.\nBe specific. If you ultimately agree after analysis, say so and explain why.\n\nCounter-analysis:\"\"\"\n\n# Pattern: rotate the adversarial role across agents each round\ndef assign_roles(n_agents: int, round_num: int) -> list[str]:\n    roles = [\"standard\"] * n_agents\n    adversarial_idx = round_num % n_agents\n    roles[adversarial_idx] = \"adversarial\"\n    return roles\n```\n\n---\n\n## Compute budget guide\n\n```\nN agents × R rounds = N×R LLM calls, each with growing context\n5 agents × 5 rounds = 25 calls at increasing context size\nCost per question can exceed 10× single CoT call\n\nPractical recommendations:\n  Low-stakes tasks:    3 agents × 2 rounds  (6 calls)  — diminishing returns past round 2\n  High-stakes tasks:   5 agents × 3 rounds  (15 calls)\n  Research synthesis:  3 heterogeneous models × 3 rounds\n  \nHeterogeneous = different model families (Claude + GPT + Llama)\n→ errors don't correlate → more diverse starting points → better convergence\n```\n\n---\n\n## Topic drift mitigation\n\n```python\ndef inject_question_every_round(question: str, prior_answers: list[str]) -> str:\n    \"\"\"Re-anchor agents to the original question every round to prevent drift.\"\"\"\n    return f\"\"\"IMPORTANT: Keep your answer focused on the original question:\n\"{question}\"\n\nDo not let discussion drift to adjacent topics.\n\nOther agents' latest answers:\n{chr(10).join(f'- {a[:300]}' for a in prior_answers)}\n\nYour focused answer to the original question:\"\"\"\n```\n\n---\n\n## Anti-Fake-Pass Checklist\n\n```\n❌ All agents from same model family → monoculture collapse; errors correlate; debate is just sampling\n❌ No adversarial role → sycophancy cascade; all agents defer to most confident answer\n❌ Round count > 3 without task-specific tuning → diminishing returns + context explosion\n❌ Topic drift across rounds → re-inject the original question every round\n❌ Weak model in the pool → drags consensus toward its wrong answer (Du et al. \"MAD\" follow-up)\n❌ No majority vote normalization → agents phrase the same answer differently; vote splits\n```\n","# \u002Fmulti-agent-debate\n\n## When to Use\n\n- High-stakes reasoning where a single LLM gives confidently wrong answers\n- Reducing correlated errors: one model family has systematic bias in a domain\n- Factual tasks where self-consistency sampling has plateaued (debate breaks the saturation)\n- Research synthesis, code review, or multi-perspective analysis\n\n## Do NOT use for\n\n- Tasks with a single deterministic answer verifiable by tool (use CRITIC in [[self-refine-critic]])\n- Latency-sensitive workloads (N × R LLM calls = N×R cost)\n- Tasks where any model in the pool is clearly much weaker than others (weak models drag consensus)\n\n---\n\n## Du et al. 2023 algorithm (ICML 2024)\n\n```\nRound 0 (initial):  each agent i produces answer_i independently\nRound r (debate):   each agent i sees all agents' round r-1 answers\n                    → produces updated answer_i conditioned on the group\nFinal:              majority-vote across agents' final answers\n```\n\nTwo independent contributions to quality (from paper ablations):\n- **Multiple agents** (1 round, majority vote): beats single-agent; plateaus\n- **Multiple rounds** (1 agent, self-reflection): barely helps alone\n- **Both together**: produces the big accuracy jumps\n\n---\n\n## Debate loop\n\n```python\nfrom typing import Callable\n\ndef debate(\n    question: str,\n    agents: list[Callable[[str, list[str] | None], str]],\n    rounds: int = 3,\n) -> str:\n    \"\"\"\n    agents: list of callables (question, prior_answers) -> answer\n    Returns majority-voted final answer.\n    \"\"\"\n    # Round 0: independent proposals\n    answers: list[str] = [agent(question, None) for agent in agents]\n\n    # Rounds 1..R: each agent sees all others' prior answers\n    for r in range(1, rounds):\n        new_answers: list[str] = []\n        for i, agent in enumerate(agents):\n            others = [answers[j] for j in range(len(agents)) if j != i]\n            updated = agent(question, others)\n            new_answers.append(updated)\n        answers = new_answers\n\n    return majority_vote(answers)\n\ndef majority_vote(answers: list[str]) -> str:\n    from collections import Counter\n    # Normalize: extract final \"Answer: X\" from each response\n    import re\n    extracted = []\n    for a in answers:\n        m = re.search(r'(?:answer|final answer)[:\\s]+(.+?)(?:\\.|$)', a, re.IGNORECASE)\n        extracted.append(m.group(1).strip() if m else a.strip())\n    counts = Counter(extracted)\n    return counts.most_common(1)[0][0]\n```\n\n---\n\n## Agent prompt with debate context\n\n```python\ndef build_debate_prompt(question: str, prior_answers: list[str] | None) -> str:\n    if not prior_answers:\n        return f\"\"\"Answer the following question. Show your reasoning.\n\nQuestion: {question}\n\nAnswer:\"\"\"\n\n    others_block = \"\\n\\n\".join(\n        f\"Agent {i+1}'s answer:\\n{ans}\"\n        for i, ans in enumerate(prior_answers)\n    )\n    return f\"\"\"Answer the following question. You have seen other agents' answers below.\nConsider their reasoning critically. You may update your answer or maintain it with justification.\n\nQuestion: {question}\n\nOther agents' answers:\n{others_block}\n\nYour updated answer (with reasoning):\"\"\"\n```\n\n---\n\n## Sycophancy cascade prevention\n\n```python\ndef build_adversarial_debate_prompt(question: str, prior_answers: list[str]) -> str:\n    \"\"\"Force at least one agent to argue the counter-position.\"\"\"\n    others_block = \"\\n\\n\".join(f\"Agent {i+1}: {a}\" for i, a in enumerate(prior_answers))\n    return f\"\"\"Question: {question}\n\nOther agents have proposed the following answers:\n{others_block}\n\nYour role: identify the weakest point in the most popular answer and argue for an alternative.\nBe specific. If you ultimately agree after analysis, say so and explain why.\n\nCounter-analysis:\"\"\"\n\n# Pattern: rotate the adversarial role across agents each round\ndef assign_roles(n_agents: int, round_num: int) -> list[str]:\n    roles = [\"standard\"] * n_agents\n    adversarial_idx = round_num % n_agents\n    roles[adversarial_idx] = \"adversarial\"\n    return roles\n```\n\n---\n\n## Compute budget guide\n\n```\nN agents × R rounds = N×R LLM calls, each with growing context\n5 agents × 5 rounds = 25 calls at increasing context size\nCost per question can exceed 10× single CoT call\n\nPractical recommendations:\n  Low-stakes tasks:    3 agents × 2 rounds  (6 calls)  — diminishing returns past round 2\n  High-stakes tasks:   5 agents × 3 rounds  (15 calls)\n  Research synthesis:  3 heterogeneous models × 3 rounds\n  \nHeterogeneous = different model families (Claude + GPT + Llama)\n→ errors don't correlate → more diverse starting points → better convergence\n```\n\n---\n\n## Topic drift mitigation\n\n```python\ndef inject_question_every_round(question: str, prior_answers: list[str]) -> str:\n    \"\"\"Re-anchor agents to the original question every round to prevent drift.\"\"\"\n    return f\"\"\"IMPORTANT: Keep your answer focused on the original question:\n\"{question}\"\n\nDo not let discussion drift to adjacent topics.\n\nOther agents' latest answers:\n{chr(10).join(f'- {a[:300]}' for a in prior_answers)}\n\nYour focused answer to the original question:\"\"\"\n```\n\n---\n\n## Anti-Fake-Pass Checklist\n\n```\n❌ All agents from same model family → monoculture collapse; errors correlate; debate is just sampling\n❌ No adversarial role → sycophancy cascade; all agents defer to most confident answer\n❌ Round count > 3 without task-specific tuning → diminishing returns + context explosion\n❌ Topic drift across rounds → re-inject the original question every round\n❌ Weak model in the pool → drags consensus toward its wrong answer (Du et al. \"MAD\" follow-up)\n❌ No majority vote normalization → agents phrase the same answer differently; vote splits\n```\n",{},"main","481dcd7dd2e288b9d627a2bd017db70e64e6c6fa","core\u002Fskills\u002Fmulti-agent-debate",2,"https:\u002F\u002Fgithub.com\u002FHoronomos-laps\u002FYana-AI","https:\u002F\u002Fyana.vutam.link\u002F","https:\u002F\u002Fgithub.com\u002FHoronomos-laps","https:\u002F\u002Fyana.vutam.link","An open-source governed runtime & execution environment for multi-agent AI systems.","2026-11-14 13:24:09",{"source":7,"skillName":5,"commands":34,"prompt":44,"zipAvailable":45},[35,38,41],{"tool":36,"command":37},"npx","npx skills add Horonomos-laps\u002FYana-AI --skill multi-agent-debate",{"tool":39,"command":40},"bunx","bunx skills add Horonomos-laps\u002FYana-AI --skill multi-agent-debate",{"tool":42,"command":43},"pnpm","pnpm dlx skills add Horonomos-laps\u002FYana-AI --skill multi-agent-debate","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FHoronomos-laps\u002FYana-AI\u002Fblob\u002Fmain\u002Fcore\u002Fskills\u002Fmulti-agent-debate\u002FSKILL.md",true,[],[48,56,63,71,79,87],{"skillNo":49,"name":50,"nameZh":6,"description":51,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":52,"licenseFileUrl":6,"sourceHtmlUrl":53,"categories":54,"keywords":55,"publishedAt":18},"01M735GKET5YVYKAY62W2P5YXB","openai--sentry--sentry","Use when the user asks to inspect Sentry issues or events, summarize recent production errors, or pull basic Sentry health data via the Sentry API; perform read-only queries with the bundled script and require `SENTRY_AUTH_TOKEN`.","MIT","https:\u002F\u002Fgithub.com\u002FHoronomos-laps\u002FYana-AI\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fopenai--sentry--sentry\u002FSKILL.md",[],[],{"skillNo":57,"name":58,"nameZh":6,"description":59,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":60,"categories":61,"keywords":62,"publishedAt":18},"01M73DCAP83T17HY7SMGK7MGFE","ito-basket-compare","Compare Itô prediction-market baskets against a user's knowledge base, portfolio notes, financial context, watchlist, or research thesis. Use for read-only basket comparison and gap analysis without investment advice or live trading.","https:\u002F\u002Fgithub.com\u002FHoronomos-laps\u002FYana-AI\u002Fblob\u002Fmain\u002Fcore\u002Fskills\u002Fito-basket-compare\u002FSKILL.md",[],[],{"skillNo":64,"name":65,"nameZh":6,"description":66,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":67,"sourceHtmlUrl":68,"categories":69,"keywords":70,"publishedAt":18},"01M734CYEMZFDR3KH6SW0JP8TD","implementing-network-traffic-baselining","Build network traffic baselines from NetFlow\u002FIPFIX data using Python pandas for statistical analysis, z-score anomaly detection, and hourly\u002Fdaily traffic pattern profiling","https:\u002F\u002Fgithub.com\u002Fyanacuti1121\u002FYana-AI\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fimplementing-network-traffic-baselining\u002FLICENSE","https:\u002F\u002Fgithub.com\u002FHoronomos-laps\u002FYana-AI\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fimplementing-network-traffic-baselining\u002FSKILL.md",[],[],{"skillNo":72,"name":73,"nameZh":6,"description":74,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":75,"sourceHtmlUrl":76,"categories":77,"keywords":78,"publishedAt":18},"01M73ASZ6NJRJM44Z2P84NGQ3Q","detecting-evasion-techniques-in-endpoint-logs","Detects defense evasion techniques used by adversaries in endpoint logs including log tampering, timestomping, process injection, and security tool disabling. Use when investigating suspicious endpoint behavior, building detection rules for evasion tactics, or conducting threat hunting for stealthy adversary activity. Activates for requests involving evasion detection, defense evasion analysis, log tampering detection, or MITRE ATT&CK TA0005.","https:\u002F\u002Fgithub.com\u002Fyanacuti1121\u002FYana-AI\u002Fblob\u002Fmain\u002Fcore\u002Fskills\u002Fdetecting-evasion-techniques-in-endpoint-logs\u002FLICENSE","https:\u002F\u002Fgithub.com\u002FHoronomos-laps\u002FYana-AI\u002Fblob\u002Fmain\u002Fcore\u002Fskills\u002Fdetecting-evasion-techniques-in-endpoint-logs\u002FSKILL.md",[],[],{"skillNo":80,"name":81,"nameZh":6,"description":82,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":11,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":83,"sourceHtmlUrl":84,"categories":85,"keywords":86,"publishedAt":18},"01M734AFNXD06Z78QRK0VKGT7Q","implementing-microsegmentation-with-guardicore","Implementing microsegmentation using Akamai Guardicore Segmentation to map application dependencies, create granular network policies, visualize east-west traffic flows, and enforce least-privilege communication between workloads across data centers and cloud.","https:\u002F\u002Fgithub.com\u002Fyanacuti1121\u002FYana-AI\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fimplementing-microsegmentation-with-guardicore\u002FLICENSE","https:\u002F\u002Fgithub.com\u002FHoronomos-laps\u002FYana-AI\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fimplementing-microsegmentation-with-guardicore\u002FSKILL.md",[],[],{"skillNo":88,"name":89,"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":90,"categories":91,"keywords":92,"publishedAt":18},"01M736TK27QMCXK0DVCH25SJN5","ramdisk-tmpfs-patterns","https:\u002F\u002Fgithub.com\u002FHoronomos-laps\u002FYana-AI\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Framdisk-tmpfs-patterns\u002FSKILL.md",[],[]]