Is there a difference between CLAUDE.md and AGENTS.md?

CLAUDE.md is specific to Anthropic's Claude Code, while AGENTS.md is a cross-tool standard. This skill can manage both and create symlinks for compatibility.

Does this skill support large monorepos?

Yes, it features a parallel exploration mode that uses subagents to analyze subsystems simultaneously, making it suitable for codebases exceeding 500k tokens.

What is an Intent Layer in the context of Claude Code?

An Intent Layer is a set of hierarchical AGENTS.md files that provide high-level context, architectural rules, and pitfalls to help AI agents navigate codebases more effectively.

How does git history mining improve AI performance?

By analyzing past bug fixes and reverts, the skill identifies non-obvious pitfalls and anti-patterns, pre-populating documentation with 'lessons learned' that prevent AI hallucinations.

How does the 'learning loop' feature work?

Whenever a non-obvious error is fixed, the skill can automatically append the solution and the 'gotcha' to the nearest AGENTS.md file to ensure the AI doesn't repeat the mistake.

Intent Layer Setup & Maintenance

Name: Intent Layer Setup & Maintenance
Author: orban

byorban

•

学習とドキュメンテーション

Establishes a hierarchical documentation framework using AGENTS.md files to help AI agents navigate and understand complex codebases like senior engineers.

The Intent Layer skill automates the creation and maintenance of hierarchical context files (AGENTS.md or CLAUDE.md) that provide AI agents with high-level architectural guidance, pitfalls, and domain-specific knowledge. By mining git history and PR reviews, it captures tribal knowledge often lost in standard documentation, enabling AI tools to perform with the context of a senior engineer. It includes specialized scripts for state detection, token estimation, and automated validation to ensure documentation remains concise, accurate, and within context window limits.

主な機能

01Real-time learning loops that capture non-obvious bug fixes as they occur

023 GitHub stars

03Automated git history and PR review mining to identify pitfalls and anti-patterns

04Automated validation scripts to prevent absolute paths and token bloat

05Parallel subagent exploration for rapid documentation of large-scale codebases

06Hierarchical AGENTS.md infrastructure for multi-level codebase context

ユースケース

01Scaffolding AI-friendly documentation for new projects to ensure context efficiency

02Capturing institutional knowledge from commit history and code reviews into actionable prompts

03Onboarding AI agents to legacy codebases with complex architectural nuances

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add orban/intent-layer intent-layer

For use in Claude.ai and ChatGPT

Download Skill