메뉴
Orchestrate multiple frontier LLMs (Claude, GPT-5.1, Gemini 3.0 Pro, Perplexity Sonar, Grok 4.1) for comprehensive research using LLM Council pattern with peer review and synthesis
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
직접 명령은 검토 Prompt를 거치지 않습니다. 실행하기 전에 소스를 확인하세요.
npx skills add https://github.com/krishagel/geoffrey --skill multi-model-research명령은 한 줄로 유지됩니다. 복사하기 전에 가로로 스크롤해 전체 내용을 확인하세요.
로컬 사본을 원하시나요? SkillsMP에서 현재 제공할 수 있는 파일을 다운로드하세요.
SOC 직업 분류 기준
SKILL.md 표시 중
| name | multi-model-research |
| description | Orchestrate multiple frontier LLMs (Claude, GPT-5.1, Gemini 3.0 Pro, Perplexity Sonar, Grok 4.1) for comprehensive research using LLM Council pattern with peer review and synthesis |
| triggers | ["deep dive","research council","multi-model research","comprehensive research","council research"] |
| allowed-tools | Bash, Read, mcp__obsidian-vault__create_vault_file |
| version | 0.1.0 |
Implements Karpathy's LLM Council pattern for superior research through parallel queries, peer review, and chairman synthesis.
Geoffrey/Claude (Native Council Member):
research.py)Python External API Orchestrator:
Use multi-model research when:
Simple Mode (Perplexity only):
Council Mode (Full council):
User: "What are the latest developments in quantum computing?"
↓
I decide: Simple query (factual, current)
↓
I call: uv run scripts/research.py --query "..." --models perplexity
↓
I read: JSON response from Perplexity
↓
I format: Markdown report with citations
↓
I save: To Obsidian Geoffrey/Research folder
↓
I return: Summary to user with Obsidian link
User: "Compare the AI strategies of OpenAI, Anthropic, and Google"
↓
I decide: Council query (comparative, complex)
↓
I call: uv run scripts/research.py --query "..." --models gpt,gemini,perplexity,grok
↓
I read: JSON with all external responses
↓
I provide: My own (Claude) research response
↓
I conduct: Peer review (each model ranks others)
↓
I request: GPT-5.1 chairman synthesis
↓
I format: Comprehensive markdown report
↓
I save: To Obsidian Geoffrey/Research folder
↓
I return: Summary with Obsidian link
All research reports saved to Obsidian include:
Citations use numeric format: [1], [2], etc.
Python Script:
cd skills/multi-model-research
uv run scripts/research.py --query "Your question" --models perplexity --output /tmp/responses.json
Config:
config.yaml - Model settings, routing rulesprompts/system_prompts.yaml - Per-model system promptsprompts/peer_review.md - Peer review templateprompts/chairman_synthesis.md - GPT-5.1 synthesis templateDependencies:
API Keys Required:
All keys configured in ~/.env file.
Simple Research:
User: "What is RAG in AI?"
I route to: Simple mode (Perplexity)
Output: Concise explanation with current examples and citations
Time: ~10 seconds
Council Research:
User: "Compare serverless vs containers for production ML workloads"
I route to: Council mode (all 4 external + me)
Process:
1. GPT-5.1: Provides comprehensive technical comparison
2. Gemini 3.0: Analyzes cost and performance trade-offs
3. Perplexity: Current industry trends and case studies
4. Grok 4.1: Developer sentiment from X/Twitter
5. Claude (me): Synthesize with nuanced analysis
6. Peer review: Each model ranks others
7. GPT-5.1 (chairman): Final synthesis
Output: Multi-perspective analysis with citations
Time: ~60 seconds
This skill implements Karpathy's LLM Council pattern released November 22, 2025.