Agents Md Generator

Agents.md Generator

This skill enables the core "Progressive Disclosure" strategy required for Agent-First development environments. Feeding monolithic documentation to AI coding agents results in context window exhaustion and hallucination. Instead, you must generate a lightweight AGENTS.md file (maximum ~100 lines) at the root of the repository. This file serves strictly as a Table of Contents (TOC) mapping directories to domains, teaching agents where to look next without giving them all knowledge up front.

Input

• A complete scan of the docs/, .agent/, and src/ directories.
• Any top-level structural files like ARCHITECTURE.md or PRODUCT_VISION.md.

Output

• A single AGENTS.md file placed at the root of the project repository.

Process

Phase 1: Repository Discovery

1. Call list_dir on docs/, docs/design, docs/exec-plans, and any .agent directories.
2. Identify core domain boundaries (e.g., API boundaries, UI boundaries).
3. Identify where key development plans and issue trackers live.

Phase 2: Index Generation

1. Map out the AGENTS.md structure in a hierarchical bullet-point tree.
2. For each key component, list the absolute or relative path to its source-of-truth markdown document.
3. Group paths logically (e.g., "Architecture", "Current Plans", "Frontend Guidelines").

Phase 3: File Creation

1. Write the content to ./AGENTS.md.
2. Ensure the top of the file explicitly instructs other reading agents: "Do not guess architecture. Read this TOC to find the correct deep-dive .md file, and use tools to read it before coding."

Rules

1. NEVER exceed 150 lines in AGENTS.md. It must remain lightweight.
2. NEVER summarize the actual rules or architectural choices inside this file. You must only provide the path to the file containing those choices.
3. ALWAYS use clear absolute paths or precise relative paths from root.
4. ALWAYS run this skill if the directory structure of the project's documentation changes significantly.