Menu
Multi-model consensus using Karpathy LLM Council pattern for critical decisions
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Baseado na classificação ocupacional SOC
| name | llm-council |
| description | Multi-model consensus using Karpathy LLM Council pattern for critical decisions |
| allowed-tools | Bash, Read, Write, TodoWrite |
Before writing ANY code, you MUST check:
.claude/library/catalog.json.claude/docs/inventories/LIBRARY-PATTERNS-GUIDE.mdD:\Projects\*| Match | Action |
|---|---|
| Library >90% | REUSE directly |
| Library 70-90% | ADAPT minimally |
| Pattern exists | FOLLOW pattern |
| In project | EXTRACT |
| No match | BUILD (add to library after) |
Run 3-stage multi-model consensus for critical decisions where:
STAGE 1: COLLECT
+---> Claude ---> Response A
|
Query --+---> Gemini ---> Response B
|
+---> Codex ----> Response C
STAGE 2: RANK
Each model reviews others (anonymized)
Produces rankings with rationale
STAGE 3: SYNTHESIZE
Chairman aggregates rankings
Produces final answer with consensus score
/llm-council "Should we use microservices or monolith for this system?"
/llm-council "Which auth approach is best?" --threshold 0.75
/llm-council "Architecture decision" --chairman gemini
bash scripts/multi-model/llm-council.sh "<query>" "<threshold>" "<chairman>"
| Parameter | Default | Description |
|---|---|---|
| threshold | 0.67 | Minimum consensus score |
| chairman | claude | Model that synthesizes final answer |
| models | [claude, gemini, codex] | Participating models |
Results stored to Memory-MCP:
multi-model/council/decisions/{query_id}{
"query": "Original question",
"final_answer": {
"synthesis": "Combined answer...",
"chairman": "claude"
},
"consensus_score": 0.85,
"responses": {
"claude": "...",
"gemini": "...",
"codex": "..."
},
"rankings": [
{"model": "A", "rank": 1, "rationale": "..."}
]
}
const decision = await runCouncil(
"Microservices vs Monolith for our scale?",
{ threshold: 0.75 }
);
if (decision.consensus_score >= 0.75) {
proceed(decision.final_answer);
} else {
escalateToHuman(decision);
}
const assessment = await runCouncil(
"Is this authentication approach secure?",
{ threshold: 0.80 }
);
// Higher threshold for security decisions
Complex browser automation workflow using claude-in-chrome MCP with mandatory sequential-thinking planning. Use when automating multi-step web interactions, form filling, navigation sequences, or web scraping.
End-to-end testing workflow for validating complete user journeys through web applications using claude-in-chrome MCP. Specializes in test assertions, suite organization, evidence collection, and pass/fail reporting.
Screenshot-based visual comparison and regression testing using claude-in-chrome MCP. Captures, compares, and validates UI states to detect layout shifts, visual bugs, and design regressions across viewports.
Structured data extraction from web pages using claude-in-chrome MCP with sequential-thinking planning. Focus on READ operations, data transformation, and pagination handling for multi-page extraction.
Use when conducting comprehensive code review for pull requests across multiple quality dimensions. Orchestrates 12-15 specialized reviewer agents across 4 phases using star topology coordination. Covers automated checks, parallel specialized reviews (quality, security, performance, architecture, documentation), integration analysis, and final merge recommendation in a 4-hour workflow.
Create Claude Code hooks with proper schemas, RBAC integration, and performance requirements. Use when implementing PreToolUse, PostToolUse, SessionStart, or any of the 10 hook event types for automation, validation, or security enforcement.