What is the benefit of 'Negative Instructions'?

Explicit 'DO NOT' sections in AGENTS.md are highly effective at preventing common agent mistakes, such as using banned libraries, including console logs, or ignoring type safety.

Why use AGENTS.md instead of standard skills?

Research indicates that AGENTS.md files achieve a 100% pass rate because they are always loaded in context, whereas on-demand skills have roughly a 53% invocation reliability due to trigger instability.

How does the compression feature work?

The skill provides patterns for 'Compressed Index Structures' that can reduce 40KB of documentation to 8KB using index maps, allowing the agent to retrieve specific info without bloating the context window.

Does this work with monorepos?

Yes, it supports hierarchical placement where a root AGENTS.md defines global rules and subdirectory-specific AGENTS.md files define rules for individual apps or packages.

Agents.md Pattern Architect

Name: Agents.md Pattern Architect
Author: dding-g

bydding-g

0•

Productivity & Workflow

Optimizes AI agent performance by implementing standardized AGENTS.md context files for reliable, persistent codebase guidance.

The agents-md skill provides specialized patterns for creating and maintaining AGENTS.md files, which serve as persistent, high-reliability context for AI agents working within your codebase. By moving beyond on-demand skill invocation—which can suffer from trigger instability—these patterns ensure that critical project rules, tech stacks, and architectural decisions are always loaded into the agent's context window. The skill includes advanced strategies for hierarchical context placement, token-saving documentation compression, and retrieval-led reasoning to ensure Claude prioritizes your project's specific requirements over its general pre-trained knowledge.

Key Features

01Standardized AGENTS.md file structure for tech stacks and architecture

02Token-saving compression patterns for large documentation files

030 GitHub stars

04Retrieval-led reasoning strategies to prioritize local project documentation

05Negative instruction (DO NOT) frameworks to prevent recurring agent mistakes

06Hierarchical context inheritance for monorepos and sub-projects

Use Cases

01Standardizing coding conventions across large teams using Claude for development

02Ensuring AI agents strictly follow specific architecture patterns like FSD or Server Components

03Optimizing context window usage by compressing project docs into efficient index maps

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add dding-g/ddingg-claude-marketplace agents-md

For use in Claude.ai and ChatGPT

Download Skill