By name: feynman_simulator description: Simulates the Feynman Technique by instantiating a multi-agent debate loop with Playwright. metadata: openclaw: emoji: 🧑🏫 display_name: 費曼學習法模擬 state_tracking: phases:
- p01_feynman_debate
- p02_debate_synthesis labels: p01_feynman_debate: P1 (Debate) p02_debate_synthesis: P2 (Synthesis) io_contracts: consumes:
- text/markdown produces:
- text/markdown
Feynman Simulator
This skill simulates the Feynman Technique by instantiating a multi-agent debate loop. It uses Playwright to bypass login walls, placing a Student Agent (local Ollama) against a Tutor Agent (Google Gemini).
Phases
- Phase 1: AI Debate (
p01_ai_debate.py): A multi-turn Socratic debate. The Student proposes explanations, and the Tutor probes for weaknesses, blind spots, and logical inconsistencies. - Phase 2: Summarize (
p02_summarize.py): Condenses the debate findings into a final refined document.
Run Commands
cd openclaw-sandbox
python3 skills/feynman_simulator/scripts/run_all.py --process-all
Config Pointers
config/config.yaml: Sets model choices — primary:deepseek-r1:8b(CoT reasoning for Socratic debate), fallback:qwen3:8b. Also configures debate rounds limit.- Playwright persistent context is used for Gemini interaction.
Safety & Hallucination Defenses
- The debate loop uses strict constraints to prevent LLMs from hallucinating external facts. All generated explanations must be grounded in the context.