This project provides a system prompt designed to eliminate verbose, corporate-sounding, or "AI-slop" from Large Language Model (LLM) outputs. It aims to transform lengthy responses into direct, informative, and concise answers, benefiting users who require efficient information retrieval across various LLMs like GPT, Gemini, and LLaMA. The primary advantage is a significant reduction in response length while preserving essential information, as demonstrated by substantial character reductions in tests.

How It Works

The core of talk-normal is a single, carefully crafted system prompt. This prompt guides LLMs to avoid filler phrases, unnecessary explanations, and redundant language. It prioritizes directness and clarity, ensuring that the output is focused and easy to digest. This approach is advantageous because it can be applied universally across different LLM architectures without requiring model retraining, and it has been empirically validated to drastically shorten responses.

Quick Start & Requirements

Installation can be achieved via several methods:

• OpenClaw: Paste the GitHub link into the OpenClaw chat.
• ClawHub: Run clawhub install talk-normal && bash skills/talk-normal/install.sh.
• Manual: git clone https://github.com/hexiecs/talk-normal.git && cd talk-normal && bash install.sh.

All methods execute an install.sh script that auto-detects AGENTS.md and injects the prompt rules. The installer is idempotent. Uninstall is performed via bash install.sh --uninstall. A new conversation is required for changes to take effect. For OpenAI API usage, copy prompt.md into the system prompt field. For ChatGPT custom instructions, use the compressed prompt-chatgpt.md due to character limits.

Highlighted Details

• Performance: Achieves an average reduction of 73% for GPT-4o-mini and 72% for GPT-5.4, with specific prompts showing reductions up to 89%.
• Real-world Application: Demonstrated effectiveness on complex financial market analysis commentary, significantly shortening verbose explanations.
• Iterative Improvement: Individual rules are continuously refined against real LLM outputs, with regressions tracked in regressions/ to ensure ongoing effectiveness.

Maintenance & Community

Contributions are welcome and can be submitted via CONTRIBUTING.md and by opening rule suggestions in Issues. The project actively iterates on its rules, tracking regressions to maintain high performance.

Licensing & Compatibility

The project is released under the MIT License. It is compatible with various LLM interfaces including OpenClaw, ClawHub, and direct API calls using prompt.md. A specialized version, prompt-chatgpt.md, is provided for ChatGPT's custom instructions field due to its character limitations.

Limitations & Caveats

The installation script modifies existing files, specifically AGENTS.md. Users integrating with ChatGPT must use the compressed prompt-chatgpt.md due to the 1500-character limit of the custom instructions field. The effectiveness is primarily measured by character reduction, and users should verify that all necessary information is retained for their specific use cases.