Krishnagangwal
Featured
CS-Fundamentals
A curated collection of Computer Science fundamentals (PDFs, notes, cheatsheets, interview question banks) for placement preparation, covering seven core subjects plus general resources.
claude-fable-5-system-prompt-clean
A token-efficient system prompt distilled from the leaked Claude Fable 5 architecture, reducing overhead from ~30,000 to ~500 tokens while preserving autonomous reasoning and anti-chatbot behaviors.
Stars
237
7-day growth
No data
Forks
48
Open issues
0
License
MIT
Last updated
2026-07-19
Why it is worth attention
It solves the bloat and incompatibility of the leaked Anthropic prompt, making it usable across ChatGPT, Gemini, and Claude with minimal token waste and preserved core executive capabilities.
Who it is for
- Developers building LLM-powered agents
- Prompt engineers optimizing system prompts
- Researchers studying reasoning architectures
- Advanced users of ChatGPT Plus or Gemini Advanced
Use cases
- Deploying autonomous multi-turn reasoning in non-Claude models
- Reducing context window usage while maintaining high-quality outputs
- Implementing anti-chatbot prose and structural re-deconstruction in LLM interactions
- Creating custom GPTs or Gems with improved executive behavior
Strengths
- Massive token savings (30k to 500 tokens) reducing latency and context degradation
- Universal compatibility across major LLM platforms (OpenAI, Google, Anthropic)
- Preserves key behavioral features: Anti-Chatbot Prose, Structural Re-Deconstruction, No Thought Narration, Zero-Engagement Traps
- Open-source with MIT license and community contributions
Considerations
- Derived from a leaked prompt, which may raise ethical or legal concerns
- Compatibility claims are limited to specific model versions (ChatGPT 5.5/5.6, Gemini 3.1 Pro/3.5 Flash, Claude)
- May not work identically on all frontier models without further tuning for edge cases
README quick start
Universal Fable 5 Engine System Prompt
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.
📊 Key Enhancements
- 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 (``) 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.
🛠️ How to Deploy
For Google AI Studio / OpenAI Developer Platform:
Paste the raw text from system_prompt.md directly into the System Instructions wrapper of your model configuration.
For Consumer Interfaces (ChatGPT Plus / Gemini Advanced):
- 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.
🧠 Core Architectural Levers Preserved
- 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!").
🤝 Contributing
Found a edge case where another frontier model mi
Description
he optimized, token-efficient version of the leaked Claude Fable 5 / Mythos 5 system prompt. Re-engineered into clean Markdown for universal execution on Gemini 3.1 Pro, ChatGPT 5.6, and advanced LLM agents.
Related repositories
Similar projects matched by category, topics, and programming language.
xuchonglang
Featured
investing-for-beginners
A structured investing guide for Chinese beginners covering US stocks, options, and cryptocurrency, with focus on foundational concepts and risk awareness.
Liunian06
Featured
ko-lesson
ko-lesson is a Codex skill that transforms course materials into an Obsidian-based learning system with feedback loops, revision tracking, and exam cycles.