Released my global AGENTS.md / CLAUDE.md for more reliable coding agent work, and WRITING.md rules for cleaner AI text – in 3 sizes, down to a 155-word section

Enhancing AI-Driven Development and Writing: Introducing My Comprehensive Rulesets

In the rapidly evolving landscape of AI and language models, maintaining reliable and effective outputs remains a significant challenge. As someone who frequently utilizes coding agents and collaborates with large language models (LLMs) for writing, I encountered recurring issues: agents misunderstanding repositories, modifying unintended code, and reporting completion without proper verification. Similarly, AI-generated text often suffers from over-polishing, generic phrasing, and superficiality, hampering clarity and authenticity.

To address these problems, I developed a set of structured guidelines that serve as best practices for AI agents and writers alike. These rules help enforce better verification, reduce unnecessary modifications, and produce cleaner, more trustworthy outputs.

My efforts resulted in two dedicated repositories:

AGENTS.md / CLAUDE.md: Global Instructions for Coding Agents
This document provides comprehensive directives for AI coding agents, emphasizing evidence-based actions, minimal scope changes, and explicit verification. It promotes efficient collaboration through parallel work and subagents, ensuring more reliable and maintainable code generation.

WRITING.md: Clear and Authentic AI Text Guidelines
Designed to curb common AI writing pitfalls, this ruleset discourages fillers, generic phrasing, and over-structured responses. It comes in three versions to suit different needs:
– Full ruleset (~3900 words) for detailed guidance
– Compact version (~1000 words) suitable for agent instructions and quick chats
– Mini version (~155 words) as a snippet within other instruction files

Both resources are publicly available and adaptable. Feel free to use, modify, or challenge these rules based on your workflows. I’ve found them immensely helpful, and I’m eager to see how they perform in diverse setups. Your feedback and experiences are welcome to refine these guidelines further.