[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M71374J9ATBFYTYW386GQCSN":3,"skill-same:01M71374J9ATBFYTYW386GQCSN":69},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":23,"publishedAt":39,"linkOnly":40,"markdownContent":41,"markdownBody":42,"frontMatter":43,"refName":46,"commitSha":47,"skillDirectory":48,"forksCount":49,"repositoryHtmlUrl":50,"repositoryHomepageUrl":6,"ownerHtmlUrl":51,"ownerBlogUrl":52,"ownerDescription":53,"updatedAt":54,"install":55,"resources":68},"01M71374J9ATBFYTYW386GQCSN","santa-method",null,"Multi-agent adversarial verification with convergence loop. Two independent review agents must both pass before output ships. Use when output must clear two independent adversarial reviewers before it ships.","EthanYoQ\u002FSkill-hub","EthanYoQ","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F257484394?v=4","01M70C9QKWR1YS0S7V26DNB18H","Ethan","01M70C9QMER047R5ANZZA6KC2P",11,"MIT","https:\u002F\u002Fgithub.com\u002FEthanYoQ\u002FSkill-hub\u002Fblob\u002Fmine\u002Fskills\u002F01-agent-engineering\u002Fsanta-method\u002FSKILL.md",[18],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000023","Skill与智能体开发","SCENARIO","使用场景",[24,25,26,27,28,29,30,31,32,33,34,35,36,37,38],"santa method","loop","when","review","verification","convergence","multi-agent","use","output","with","multi","adversarial","founder","clear","it","2026-11-10 10:12:30",false,"---\nname: santa-method\ndescription: \"Multi-agent adversarial verification with convergence loop. Two independent review agents must both pass before output ships. Use when output must clear two independent adversarial reviewers before it ships.\"\nmetadata:\n  origin: \"Ronald Skelton - Founder, RapportScore.ai\"\n---\n\n# Santa Method\n\nMulti-agent adversarial verification framework. Make a list, check it twice. If it's naughty, fix it until it's nice.\n\nThe core insight: a single agent reviewing its own output shares the same biases, knowledge gaps, and systematic errors that produced the output. Two independent reviewers with no shared context break this failure mode.\n\n## When to Activate\n\nInvoke this skill when:\n- Output will be published, deployed, or consumed by end users\n- Compliance, regulatory, or brand constraints must be enforced\n- Code ships to production without human review\n- Content accuracy matters (technical docs, educational material, customer-facing copy)\n- Batch generation at scale where spot-checking misses systemic patterns\n- Hallucination risk is elevated (claims, statistics, API references, legal language)\n\nDo NOT use for internal drafts, exploratory research, or tasks with deterministic verification (use build\u002Ftest\u002Flint pipelines for those).\n\n## Architecture\n\n```\n┌─────────────┐\n│  GENERATOR   │  Phase 1: Make a List\n│  (Agent A)   │  Produce the deliverable\n└──────┬───────┘\n       │ output\n       ▼\n┌──────────────────────────────┐\n│     DUAL INDEPENDENT REVIEW   │  Phase 2: Check It Twice\n│                                │\n│  ┌───────────┐ ┌───────────┐  │  Two agents, same rubric,\n│  │ Reviewer B │ │ Reviewer C │  │  no shared context\n│  └─────┬─────┘ └─────┬─────┘  │\n│        │              │        │\n└────────┼──────────────┼────────┘\n         │              │\n         ▼              ▼\n┌──────────────────────────────┐\n│        VERDICT GATE           │  Phase 3: Naughty or Nice\n│                                │\n│  B passes AND C passes → NICE  │  Both must pass.\n│  Otherwise → NAUGHTY           │  No exceptions.\n└──────┬──────────────┬─────────┘\n       │              │\n    NICE           NAUGHTY\n       │              │\n       ▼              ▼\n   [ SHIP ]    ┌─────────────┐\n               │  FIX CYCLE   │  Phase 4: Fix Until Nice\n               │              │\n               │ iteration++  │  Collect all flags.\n               │ if i > MAX:  │  Fix all issues.\n               │   escalate   │  Re-run both reviewers.\n               │ else:        │  Loop until convergence.\n               │   goto Ph.2  │\n               └──────────────┘\n```\n\n## Phase Details\n\n### Phase 1: Make a List (Generate)\n\nExecute the primary task. No changes to your normal generation workflow. Santa Method is a post-generation verification layer, not a generation strategy.\n\n```python\n# The generator runs as normal\noutput = generate(task_spec)\n```\n\n### Phase 2: Check It Twice (Independent Dual Review)\n\nSpawn two review agents in parallel. Critical invariants:\n\n1. **Context isolation** — neither reviewer sees the other's assessment\n2. **Identical rubric** — both receive the same evaluation criteria\n3. **Same inputs** — both receive the original spec AND the generated output\n4. **Structured output** — each returns a typed verdict, not prose\n\n```python\nREVIEWER_PROMPT = \"\"\"\nYou are an independent quality reviewer. You have NOT seen any other review of this output.\n\n## Task Specification\n{task_spec}\n\n## Output Under Review\n{output}\n\n## Evaluation Rubric\n{rubric}\n\n## Instructions\nEvaluate the output against EACH rubric criterion. For each:\n- PASS: criterion fully met, no issues\n- FAIL: specific issue found (cite the exact problem)\n\nReturn your assessment as structured JSON:\n{\n  \"verdict\": \"PASS\" | \"FAIL\",\n  \"checks\": [\n    {\"criterion\": \"...\", \"result\": \"PASS|FAIL\", \"detail\": \"...\"}\n  ],\n  \"critical_issues\": [\"...\"],   \u002F\u002F blockers that must be fixed\n  \"suggestions\": [\"...\"]         \u002F\u002F non-blocking improvements\n}\n\nBe rigorous. Your job is to find problems, not to approve.\n\"\"\"\n```\n\n```python\n# Spawn reviewers in parallel (Claude Code subagents)\nreview_b = Agent(prompt=REVIEWER_PROMPT.format(...), description=\"Santa Reviewer B\")\nreview_c = Agent(prompt=REVIEWER_PROMPT.format(...), description=\"Santa Reviewer C\")\n\n# Both run concurrently — neither sees the other\n```\n\n### Rubric Design\n\nThe rubric is the most important input. Vague rubrics produce vague reviews. Every criterion must have an objective pass\u002Ffail condition.\n\n| Criterion | Pass Condition | Failure Signal |\n|-----------|---------------|----------------|\n| Factual accuracy | All claims verifiable against source material or common knowledge | Invented statistics, wrong version numbers, nonexistent APIs |\n| Hallucination-free | No fabricated entities, quotes, URLs, or references | Links to pages that don't exist, attributed quotes with no source |\n| Completeness | Every requirement in the spec is addressed | Missing sections, skipped edge cases, incomplete coverage |\n| Compliance | Passes all project-specific constraints | Banned terms used, tone violations, regulatory non-compliance |\n| Internal consistency | No contradictions within the output | Section A says X, section B says not-X |\n| Technical correctness | Code compiles\u002Fruns, algorithms are sound | Syntax errors, logic bugs, wrong complexity claims |\n\n#### Domain-Specific Rubric Extensions\n\n**Content\u002FMarketing:**\n- Brand voice adherence\n- SEO requirements met (keyword density, meta tags, structure)\n- No competitor trademark misuse\n- CTA present and correctly linked\n\n**Code:**\n- Type safety (no `any` leaks, proper null handling)\n- Error handling coverage\n- Security (no secrets in code, input validation, injection prevention)\n- Test coverage for new paths\n\n**Compliance-Sensitive (regulated, legal, financial):**\n- No outcome guarantees or unsubstantiated claims\n- Required disclaimers present\n- Approved terminology only\n- Jurisdiction-appropriate language\n\n### Phase 3: Naughty or Nice (Verdict Gate)\n\n```python\ndef santa_verdict(review_b, review_c):\n    \"\"\"Both reviewers must pass. No partial credit.\"\"\"\n    if review_b.verdict == \"PASS\" and review_c.verdict == \"PASS\":\n        return \"NICE\"  # Ship it\n\n    # Merge flags from both reviewers, deduplicate\n    all_issues = dedupe(review_b.critical_issues + review_c.critical_issues)\n    all_suggestions = dedupe(review_b.suggestions + review_c.suggestions)\n\n    return \"NAUGHTY\", all_issues, all_suggestions\n```\n\nWhy both must pass: if only one reviewer catches an issue, that issue is real. The other reviewer's blind spot is exactly the failure mode Santa Method exists to eliminate.\n\n### Phase 4: Fix Until Nice (Convergence Loop)\n\n```python\nMAX_ITERATIONS = 3\n\nfor iteration in range(MAX_ITERATIONS):\n    verdict, issues, suggestions = santa_verdict(review_b, review_c)\n\n    if verdict == \"NICE\":\n        log_santa_result(output, iteration, \"passed\")\n        return ship(output)\n\n    # Fix all critical issues (suggestions are optional)\n    output = fix_agent.execute(\n        output=output,\n        issues=issues,\n        instruction=\"Fix ONLY the flagged issues. Do not refactor or add unrequested changes.\"\n    )\n\n    # Re-run BOTH reviewers on fixed output (fresh agents, no memory of previous round)\n    review_b = Agent(prompt=REVIEWER_PROMPT.format(output=output, ...))\n    review_c = Agent(prompt=REVIEWER_PROMPT.format(output=output, ...))\n\n# Exhausted iterations — escalate\nlog_santa_result(output, MAX_ITERATIONS, \"escalated\")\nescalate_to_human(output, issues)\n```\n\nCritical: each review round uses **fresh agents**. Reviewers must not carry memory from previous rounds, as prior context creates anchoring bias.\n\n## Implementation Patterns\n\n### Pattern A: Claude Code Subagents (Recommended)\n\nSubagents provide true context isolation. Each reviewer is a separate process with no shared state.\n\n```bash\n# In a Claude Code session, use the Agent tool to spawn reviewers\n# Both agents run in parallel for speed\n```\n\n```python\n# Pseudocode for Agent tool invocation\nreviewer_b = Agent(\n    description=\"Santa Review B\",\n    prompt=f\"Review this output for quality...\\n\\nRUBRIC:\\n{rubric}\\n\\nOUTPUT:\\n{output}\"\n)\nreviewer_c = Agent(\n    description=\"Santa Review C\",\n    prompt=f\"Review this output for quality...\\n\\nRUBRIC:\\n{rubric}\\n\\nOUTPUT:\\n{output}\"\n)\n```\n\n### Pattern B: Sequential Inline (Fallback)\n\nWhen subagents aren't available, simulate isolation with explicit context resets:\n\n1. Generate output\n2. New context: \"You are Reviewer 1. Evaluate ONLY against this rubric. Find problems.\"\n3. Record findings verbatim\n4. Clear context completely\n5. New context: \"You are Reviewer 2. Evaluate ONLY against this rubric. Find problems.\"\n6. Compare both reviews, fix, repeat\n\nThe subagent pattern is strictly superior — inline simulation risks context bleed between reviewers.\n\n### Pattern C: Batch Sampling\n\nFor large batches (100+ items), full Santa on every item is cost-prohibitive. Use stratified sampling:\n\n1. Run Santa on a random sample (10-15% of batch, minimum 5 items)\n2. Categorize failures by type (hallucination, compliance, completeness, etc.)\n3. If systematic patterns emerge, apply targeted fixes to the entire batch\n4. Re-sample and re-verify the fixed batch\n5. Continue until a clean sample passes\n\n```python\nimport random\n\ndef santa_batch(items, rubric, sample_rate=0.15):\n    sample = random.sample(items, max(5, int(len(items) * sample_rate)))\n\n    for item in sample:\n        result = santa_full(item, rubric)\n        if result.verdict == \"NAUGHTY\":\n            pattern = classify_failure(result.issues)\n            items = batch_fix(items, pattern)  # Fix all items matching pattern\n            return santa_batch(items, rubric)   # Re-sample\n\n    return items  # Clean sample → ship batch\n```\n\n## Failure Modes and Mitigations\n\n| Failure Mode | Symptom | Mitigation |\n|-------------|---------|------------|\n| Infinite loop | Reviewers keep finding new issues after fixes | Max iteration cap (3). Escalate. |\n| Rubber stamping | Both reviewers pass everything | Adversarial prompt: \"Your job is to find problems, not approve.\" |\n| Subjective drift | Reviewers flag style preferences, not errors | Tight rubric with objective pass\u002Ffail criteria only |\n| Fix regression | Fixing issue A introduces issue B | Fresh reviewers each round catch regressions |\n| Reviewer agreement bias | Both reviewers miss the same thing | Mitigated by independence, not eliminated. For critical output, add a third reviewer or human spot-check. |\n| Cost explosion | Too many iterations on large outputs | Batch sampling pattern. Budget caps per verification cycle. |\n\n## Integration with Other Skills\n\n| Skill | Relationship |\n|-------|-------------|\n| Verification Loop | Use for deterministic checks (build, lint, test). Santa for semantic checks (accuracy, hallucinations). Run verification-loop first, Santa second. |\n| Eval Harness | Santa Method results feed eval metrics. Track pass@k across Santa runs to measure generator quality over time. |\n| Continuous Learning v2 | Santa findings become instincts. Repeated failures on the same criterion → learned behavior to avoid the pattern. |\n| Strategic Compact | Run Santa BEFORE compacting. Don't lose review context mid-verification. |\n\n## Metrics\n\nTrack these to measure Santa Method effectiveness:\n\n- **First-pass rate**: % of outputs that pass Santa on round 1 (target: >70%)\n- **Mean iterations to convergence**: average rounds to NICE (target: \u003C1.5)\n- **Issue taxonomy**: distribution of failure types (hallucination vs. completeness vs. compliance)\n- **Reviewer agreement**: % of issues flagged by both reviewers vs. only one (low agreement = rubric needs tightening)\n- **Escape rate**: issues found post-ship that Santa should have caught (target: 0)\n\n## Cost Analysis\n\nSanta Method costs approximately 2-3x the token cost of generation alone per verification cycle. For most high-stakes output, this is a bargain:\n\n```\nCost of Santa = (generation tokens) + 2×(review tokens per round) × (avg rounds)\nCost of NOT Santa = (reputation damage) + (correction effort) + (trust erosion)\n```\n\nFor batch operations, the sampling pattern reduces cost to ~15-20% of full verification while catching >90% of systematic issues.\n","# Santa Method\n\nMulti-agent adversarial verification framework. Make a list, check it twice. If it's naughty, fix it until it's nice.\n\nThe core insight: a single agent reviewing its own output shares the same biases, knowledge gaps, and systematic errors that produced the output. Two independent reviewers with no shared context break this failure mode.\n\n## When to Activate\n\nInvoke this skill when:\n- Output will be published, deployed, or consumed by end users\n- Compliance, regulatory, or brand constraints must be enforced\n- Code ships to production without human review\n- Content accuracy matters (technical docs, educational material, customer-facing copy)\n- Batch generation at scale where spot-checking misses systemic patterns\n- Hallucination risk is elevated (claims, statistics, API references, legal language)\n\nDo NOT use for internal drafts, exploratory research, or tasks with deterministic verification (use build\u002Ftest\u002Flint pipelines for those).\n\n## Architecture\n\n```\n┌─────────────┐\n│  GENERATOR   │  Phase 1: Make a List\n│  (Agent A)   │  Produce the deliverable\n└──────┬───────┘\n       │ output\n       ▼\n┌──────────────────────────────┐\n│     DUAL INDEPENDENT REVIEW   │  Phase 2: Check It Twice\n│                                │\n│  ┌───────────┐ ┌───────────┐  │  Two agents, same rubric,\n│  │ Reviewer B │ │ Reviewer C │  │  no shared context\n│  └─────┬─────┘ └─────┬─────┘  │\n│        │              │        │\n└────────┼──────────────┼────────┘\n         │              │\n         ▼              ▼\n┌──────────────────────────────┐\n│        VERDICT GATE           │  Phase 3: Naughty or Nice\n│                                │\n│  B passes AND C passes → NICE  │  Both must pass.\n│  Otherwise → NAUGHTY           │  No exceptions.\n└──────┬──────────────┬─────────┘\n       │              │\n    NICE           NAUGHTY\n       │              │\n       ▼              ▼\n   [ SHIP ]    ┌─────────────┐\n               │  FIX CYCLE   │  Phase 4: Fix Until Nice\n               │              │\n               │ iteration++  │  Collect all flags.\n               │ if i > MAX:  │  Fix all issues.\n               │   escalate   │  Re-run both reviewers.\n               │ else:        │  Loop until convergence.\n               │   goto Ph.2  │\n               └──────────────┘\n```\n\n## Phase Details\n\n### Phase 1: Make a List (Generate)\n\nExecute the primary task. No changes to your normal generation workflow. Santa Method is a post-generation verification layer, not a generation strategy.\n\n```python\n# The generator runs as normal\noutput = generate(task_spec)\n```\n\n### Phase 2: Check It Twice (Independent Dual Review)\n\nSpawn two review agents in parallel. Critical invariants:\n\n1. **Context isolation** — neither reviewer sees the other's assessment\n2. **Identical rubric** — both receive the same evaluation criteria\n3. **Same inputs** — both receive the original spec AND the generated output\n4. **Structured output** — each returns a typed verdict, not prose\n\n```python\nREVIEWER_PROMPT = \"\"\"\nYou are an independent quality reviewer. You have NOT seen any other review of this output.\n\n## Task Specification\n{task_spec}\n\n## Output Under Review\n{output}\n\n## Evaluation Rubric\n{rubric}\n\n## Instructions\nEvaluate the output against EACH rubric criterion. For each:\n- PASS: criterion fully met, no issues\n- FAIL: specific issue found (cite the exact problem)\n\nReturn your assessment as structured JSON:\n{\n  \"verdict\": \"PASS\" | \"FAIL\",\n  \"checks\": [\n    {\"criterion\": \"...\", \"result\": \"PASS|FAIL\", \"detail\": \"...\"}\n  ],\n  \"critical_issues\": [\"...\"],   \u002F\u002F blockers that must be fixed\n  \"suggestions\": [\"...\"]         \u002F\u002F non-blocking improvements\n}\n\nBe rigorous. Your job is to find problems, not to approve.\n\"\"\"\n```\n\n```python\n# Spawn reviewers in parallel (Claude Code subagents)\nreview_b = Agent(prompt=REVIEWER_PROMPT.format(...), description=\"Santa Reviewer B\")\nreview_c = Agent(prompt=REVIEWER_PROMPT.format(...), description=\"Santa Reviewer C\")\n\n# Both run concurrently — neither sees the other\n```\n\n### Rubric Design\n\nThe rubric is the most important input. Vague rubrics produce vague reviews. Every criterion must have an objective pass\u002Ffail condition.\n\n| Criterion | Pass Condition | Failure Signal |\n|-----------|---------------|----------------|\n| Factual accuracy | All claims verifiable against source material or common knowledge | Invented statistics, wrong version numbers, nonexistent APIs |\n| Hallucination-free | No fabricated entities, quotes, URLs, or references | Links to pages that don't exist, attributed quotes with no source |\n| Completeness | Every requirement in the spec is addressed | Missing sections, skipped edge cases, incomplete coverage |\n| Compliance | Passes all project-specific constraints | Banned terms used, tone violations, regulatory non-compliance |\n| Internal consistency | No contradictions within the output | Section A says X, section B says not-X |\n| Technical correctness | Code compiles\u002Fruns, algorithms are sound | Syntax errors, logic bugs, wrong complexity claims |\n\n#### Domain-Specific Rubric Extensions\n\n**Content\u002FMarketing:**\n- Brand voice adherence\n- SEO requirements met (keyword density, meta tags, structure)\n- No competitor trademark misuse\n- CTA present and correctly linked\n\n**Code:**\n- Type safety (no `any` leaks, proper null handling)\n- Error handling coverage\n- Security (no secrets in code, input validation, injection prevention)\n- Test coverage for new paths\n\n**Compliance-Sensitive (regulated, legal, financial):**\n- No outcome guarantees or unsubstantiated claims\n- Required disclaimers present\n- Approved terminology only\n- Jurisdiction-appropriate language\n\n### Phase 3: Naughty or Nice (Verdict Gate)\n\n```python\ndef santa_verdict(review_b, review_c):\n    \"\"\"Both reviewers must pass. No partial credit.\"\"\"\n    if review_b.verdict == \"PASS\" and review_c.verdict == \"PASS\":\n        return \"NICE\"  # Ship it\n\n    # Merge flags from both reviewers, deduplicate\n    all_issues = dedupe(review_b.critical_issues + review_c.critical_issues)\n    all_suggestions = dedupe(review_b.suggestions + review_c.suggestions)\n\n    return \"NAUGHTY\", all_issues, all_suggestions\n```\n\nWhy both must pass: if only one reviewer catches an issue, that issue is real. The other reviewer's blind spot is exactly the failure mode Santa Method exists to eliminate.\n\n### Phase 4: Fix Until Nice (Convergence Loop)\n\n```python\nMAX_ITERATIONS = 3\n\nfor iteration in range(MAX_ITERATIONS):\n    verdict, issues, suggestions = santa_verdict(review_b, review_c)\n\n    if verdict == \"NICE\":\n        log_santa_result(output, iteration, \"passed\")\n        return ship(output)\n\n    # Fix all critical issues (suggestions are optional)\n    output = fix_agent.execute(\n        output=output,\n        issues=issues,\n        instruction=\"Fix ONLY the flagged issues. Do not refactor or add unrequested changes.\"\n    )\n\n    # Re-run BOTH reviewers on fixed output (fresh agents, no memory of previous round)\n    review_b = Agent(prompt=REVIEWER_PROMPT.format(output=output, ...))\n    review_c = Agent(prompt=REVIEWER_PROMPT.format(output=output, ...))\n\n# Exhausted iterations — escalate\nlog_santa_result(output, MAX_ITERATIONS, \"escalated\")\nescalate_to_human(output, issues)\n```\n\nCritical: each review round uses **fresh agents**. Reviewers must not carry memory from previous rounds, as prior context creates anchoring bias.\n\n## Implementation Patterns\n\n### Pattern A: Claude Code Subagents (Recommended)\n\nSubagents provide true context isolation. Each reviewer is a separate process with no shared state.\n\n```bash\n# In a Claude Code session, use the Agent tool to spawn reviewers\n# Both agents run in parallel for speed\n```\n\n```python\n# Pseudocode for Agent tool invocation\nreviewer_b = Agent(\n    description=\"Santa Review B\",\n    prompt=f\"Review this output for quality...\\n\\nRUBRIC:\\n{rubric}\\n\\nOUTPUT:\\n{output}\"\n)\nreviewer_c = Agent(\n    description=\"Santa Review C\",\n    prompt=f\"Review this output for quality...\\n\\nRUBRIC:\\n{rubric}\\n\\nOUTPUT:\\n{output}\"\n)\n```\n\n### Pattern B: Sequential Inline (Fallback)\n\nWhen subagents aren't available, simulate isolation with explicit context resets:\n\n1. Generate output\n2. New context: \"You are Reviewer 1. Evaluate ONLY against this rubric. Find problems.\"\n3. Record findings verbatim\n4. Clear context completely\n5. New context: \"You are Reviewer 2. Evaluate ONLY against this rubric. Find problems.\"\n6. Compare both reviews, fix, repeat\n\nThe subagent pattern is strictly superior — inline simulation risks context bleed between reviewers.\n\n### Pattern C: Batch Sampling\n\nFor large batches (100+ items), full Santa on every item is cost-prohibitive. Use stratified sampling:\n\n1. Run Santa on a random sample (10-15% of batch, minimum 5 items)\n2. Categorize failures by type (hallucination, compliance, completeness, etc.)\n3. If systematic patterns emerge, apply targeted fixes to the entire batch\n4. Re-sample and re-verify the fixed batch\n5. Continue until a clean sample passes\n\n```python\nimport random\n\ndef santa_batch(items, rubric, sample_rate=0.15):\n    sample = random.sample(items, max(5, int(len(items) * sample_rate)))\n\n    for item in sample:\n        result = santa_full(item, rubric)\n        if result.verdict == \"NAUGHTY\":\n            pattern = classify_failure(result.issues)\n            items = batch_fix(items, pattern)  # Fix all items matching pattern\n            return santa_batch(items, rubric)   # Re-sample\n\n    return items  # Clean sample → ship batch\n```\n\n## Failure Modes and Mitigations\n\n| Failure Mode | Symptom | Mitigation |\n|-------------|---------|------------|\n| Infinite loop | Reviewers keep finding new issues after fixes | Max iteration cap (3). Escalate. |\n| Rubber stamping | Both reviewers pass everything | Adversarial prompt: \"Your job is to find problems, not approve.\" |\n| Subjective drift | Reviewers flag style preferences, not errors | Tight rubric with objective pass\u002Ffail criteria only |\n| Fix regression | Fixing issue A introduces issue B | Fresh reviewers each round catch regressions |\n| Reviewer agreement bias | Both reviewers miss the same thing | Mitigated by independence, not eliminated. For critical output, add a third reviewer or human spot-check. |\n| Cost explosion | Too many iterations on large outputs | Batch sampling pattern. Budget caps per verification cycle. |\n\n## Integration with Other Skills\n\n| Skill | Relationship |\n|-------|-------------|\n| Verification Loop | Use for deterministic checks (build, lint, test). Santa for semantic checks (accuracy, hallucinations). Run verification-loop first, Santa second. |\n| Eval Harness | Santa Method results feed eval metrics. Track pass@k across Santa runs to measure generator quality over time. |\n| Continuous Learning v2 | Santa findings become instincts. Repeated failures on the same criterion → learned behavior to avoid the pattern. |\n| Strategic Compact | Run Santa BEFORE compacting. Don't lose review context mid-verification. |\n\n## Metrics\n\nTrack these to measure Santa Method effectiveness:\n\n- **First-pass rate**: % of outputs that pass Santa on round 1 (target: >70%)\n- **Mean iterations to convergence**: average rounds to NICE (target: \u003C1.5)\n- **Issue taxonomy**: distribution of failure types (hallucination vs. completeness vs. compliance)\n- **Reviewer agreement**: % of issues flagged by both reviewers vs. only one (low agreement = rubric needs tightening)\n- **Escape rate**: issues found post-ship that Santa should have caught (target: 0)\n\n## Cost Analysis\n\nSanta Method costs approximately 2-3x the token cost of generation alone per verification cycle. For most high-stakes output, this is a bargain:\n\n```\nCost of Santa = (generation tokens) + 2×(review tokens per round) × (avg rounds)\nCost of NOT Santa = (reputation damage) + (correction effort) + (trust erosion)\n```\n\nFor batch operations, the sampling pattern reduces cost to ~15-20% of full verification while catching >90% of systematic issues.\n",{"name":5,"metadata":44,"description":7},{"origin":45},"Ronald Skelton - Founder, RapportScore.ai","mine","1e7390741dffd3d5041fca4ebc056071ee3ba137","skills\u002F01-agent-engineering\u002Fsanta-method",5,"https:\u002F\u002Fgithub.com\u002FEthanYoQ\u002FSkill-hub","https:\u002F\u002Fgithub.com\u002FEthanYoQ","www.orz.md","Part Marketer, Part AI Explorer.","2026-11-07 14:52:29",{"source":8,"skillName":5,"commands":56,"prompt":66,"zipAvailable":67},[57,60,63],{"tool":58,"command":59},"npx","npx skills add EthanYoQ\u002FSkill-hub --skill santa-method",{"tool":61,"command":62},"bunx","bunx skills add EthanYoQ\u002FSkill-hub --skill santa-method",{"tool":64,"command":65},"pnpm","pnpm dlx skills add EthanYoQ\u002FSkill-hub --skill santa-method","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FEthanYoQ\u002FSkill-hub\u002Fblob\u002Fmine\u002Fskills\u002F01-agent-engineering\u002Fsanta-method\u002FSKILL.md",true,[],[70,97,128,151,169,184],{"skillNo":71,"name":72,"nameZh":6,"description":73,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":74,"categories":75,"keywords":77,"publishedAt":39},"01M713CBDSYARG4PXFXK2R2NNN","xhs-pharma-social-listening","Run evidence-disciplined Xiaohongshu social listening for foreign pharma employee pain points. Use when the user asks to collect and analyze notes\u002Fcomments about pharma companies, aliases, roles, compliance, medical affairs, or AI topics, then produce a cited and scored sampled-insight report. Includes independent-evidence counting, MediaCrawler batch collection, and CAPTCHA\u002F461 fallback to Agent Reach\u002FOpenCLI.","https:\u002F\u002Fgithub.com\u002FEthanYoQ\u002FSkill-hub\u002Fblob\u002Fmine\u002Fskills\u002F10-business-industry\u002Fxhs-pharma-social-listening\u002FSKILL.md",[76],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},[78,79,80,81,82,83,26,84,85,86,87,88,89,90,91,92,93,94,95,96],"social","xhs pharma social listening","social listening","xhs","medical","counting","collection","batch","analyze","notes","run","opencli","fallback","captcha","agent reach","evidence","for","collect","produce",{"skillNo":98,"name":99,"nameZh":6,"description":100,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":101,"categories":102,"keywords":109,"publishedAt":39},"01M713C8Q76ZP0WXXBRAFKS6EE","healthcare-phi-compliance","Protected Health Information (PHI) and Personally Identifiable Information (PII) compliance patterns for healthcare applications. Covers data classification, access control, audit trails, encryption, and common leak vectors. Use when code touches PHI or PII in a healthcare system, or when auditing access control, audit trails, or leak vectors.","https:\u002F\u002Fgithub.com\u002FEthanYoQ\u002FSkill-hub\u002Fblob\u002Fmine\u002Fskills\u002F10-business-industry\u002Fhealthcare-phi-compliance\u002FSKILL.md",[103,104],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},{"categoryNo":105,"categoryName":106,"taxonomyType":107,"taxonomyTypeDesc":108},"01JBSEEDCAT000000000000037","律师与法务","PROFESSION","适用职业",[110,111,112,113,114,115,116,117,26,118,119,120,121,122,123,124,94,125,126,127],"healthcare","phi","compliance","healthcare phi compliance","pii","compliance patterns","information","access","encryption","code","common","audit","data","patterns","Access Control","system","applications","Auditing Access Control",{"skillNo":129,"name":130,"nameZh":6,"description":131,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":132,"categories":133,"keywords":135,"publishedAt":39},"01M713C5R2GE43X45T2K75TB76","healthcare-eval-harness","Patient safety evaluation harness for healthcare application deployments. Automated test suites for CDSS accuracy, PHI exposure, clinical workflow integrity, and integration compliance. Blocks deployments on safety failures. Use when a healthcare deployment must be gated on patient-safety tests for CDSS accuracy, PHI exposure, and workflow integrity.","https:\u002F\u002Fgithub.com\u002FEthanYoQ\u002FSkill-hub\u002Fblob\u002Fmine\u002Fskills\u002F10-business-industry\u002Fhealthcare-eval-harness\u002FSKILL.md",[134],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},[110,136,137,138,139,140,141,26,142,143,144,145,111,94,146,147,148,149,31,150],"eval","harness","healthcare eval harness","eval harness","clinical","patient","workflow","Evaluation","Safety","integrity","deployment","accuracy","tests","deployments","super",{"skillNo":152,"name":153,"nameZh":6,"description":154,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":155,"categories":156,"keywords":158,"publishedAt":39},"01M713C2J17S91VYR7A8M696VR","healthcare-emr-patterns","EMR\u002FEHR development patterns for healthcare applications. Clinical safety, encounter workflows, prescription generation, clinical decision support integration, and accessibility-first UI for medical data entry. Use when building EMR or EHR features such as encounter workflows, prescription generation, or clinical data entry UI.","https:\u002F\u002Fgithub.com\u002FEthanYoQ\u002FSkill-hub\u002Fblob\u002Fmine\u002Fskills\u002F10-business-industry\u002Fhealthcare-emr-patterns\u002FSKILL.md",[157],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},[110,159,123,160,161,82,140,162,26,163,164,165,144,122,166,167,168,94,126,31],"EMR","healthcare emr patterns","decision","workflows","support","clinical decision support","features","ui","as","development",{"skillNo":170,"name":171,"nameZh":6,"description":172,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":173,"categories":174,"keywords":176,"publishedAt":39},"01M713BZERTXV0N8X6SW9DDVWZ","healthcare-cdss-patterns","Clinical Decision Support System (CDSS) development patterns. Drug interaction checking, dose validation, clinical scoring (NEWS2, qSOFA), alert severity classification, and integration into EMR workflows. Use when building clinical decision support — drug interaction checks, dose validation, clinical scoring, or alert severity.","https:\u002F\u002Fgithub.com\u002FEthanYoQ\u002FSkill-hub\u002Fblob\u002Fmine\u002Fskills\u002F10-business-industry\u002Fhealthcare-cdss-patterns\u002FSKILL.md",[175],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},[110,123,177,161,140,159,162,26,178,179,163,164,180,168,125,181,31,182,150,183],"healthcare cdss patterns","scoring","interaction","severity","classification","alert","decision support",{"skillNo":185,"name":186,"nameZh":6,"description":187,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":188,"categories":189,"keywords":191,"publishedAt":39},"01M713BVK3PQR7V4ZKYGMWMXJD","dxy-task-clarify","用户的请求含糊、缺背景、缺资料说明、缺读者或用途时使用。先用少量带选项的追问把任务问清楚，写成任务卡，用户确认后再执行。尤其适用于医生、科研人员和药企各岗位的工作，例如「帮我分析一下这些数据」「帮我准备明天的拜访」「写个总结」「做个 PPT」「回一下这封邮件」「整理一下会议纪要」「整理一下这些文献」「帮我准备下周的讲课」「写个病例汇报」「帮我写个提示词」。用户说「先问我」「帮我理清需求」「我不知道怎么描述」「我不会问 AI」时也使用。用途、资料、读者都已说清的请求，纯翻译，以及写代码、改代码，不使用。","https:\u002F\u002Fgithub.com\u002FEthanYoQ\u002FSkill-hub\u002Fblob\u002Fmine\u002Fskills\u002F10-business-industry\u002Fdxy-task-clarify\u002FSKILL.md",[190],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},[192,193,194,195,196,197,198,199],"clarify","dxy task clarify","task","pptx","整理","翻译","数据","总结"]