A high-performance, token-efficient distillation of the leaked Claude Fable 5 / Mythos 5 system prompt architecture.
When the raw 120,000-character Anthropic system prompt was leaked, developers quickly realized it was too bloated for non-Claude models due to thousands of lines of environment-specific XML tools, safety overrides, and server-side functions.
This repository fixes that. We have stripped away the proprietary Anthropic infrastructure bloat while perfectly preserving Fable 5's autonomous reasoning, multi-turn horizon planning, self-verification loops, and distinct anti-chatbot behavioral rules.
- Massive Token Savings: Slashes prompt overhead from ~30,000 tokens down to ~500 tokens, preventing context degradation and latency lag.
- Universal Compatibility: Removed Claude-native XML wrappers (
<antml>) that break or confuse OpenAI's ChatGPT 5.5/5.6 or Google's Gemini 3.1 Pro/3.5 Flash. - Pure Operational Philosophy: Retains the exact behavioral scaffolding that gives Fable 5 its elite, senior-engineer executive capabilities.
Paste the raw text from system_prompt.md directly into the System Instructions wrapper of your model configuration.
- ChatGPT: Create a New GPT or paste the prompt text into your Custom Instructions profile settings.
- Gemini Advanced: Paste the text directly into the Gems builder system prompt panel.
- The Anti-Chatbot Prose Standard: Actively forces the model out of typical conversational patterns (walls of bullets, aggressive nesting, fake enthusiasm) and mandates clean, high-density continuous prose.
- Structural Re-Deconstruction: Prevents lazy summaries by forcing the model to systematically smash and rebuild incoming source data flows from scratch.
- No Thought Narration: Shuts off conversational meta-commentary ("Let me think about that," "Now I will compute X"), saving user tokens and execution time.
- Zero-Engagement Traps: Eliminates routine AI validation farming ("Let me know if you need more help!").
Found a edge case where another frontier model misinterprets the prompt logic? Submit a Pull Request! Let's keep this prompt optimized for upcoming model architectures.