agents-md-crafter
AGENTS.md Expert
This skill provides the standard specification, templates, and domain knowledge required to create, modify, update, and improve AGENTS.md (or similar) files for a repository. These files act as a "README for AI agents," providing the specific, detailed context agents need to work effectively.
By default, if no specific framework is requested, you should create a highly compatible AGENTS.md file based on the provided template and the official https://agents.md/ specification.
Workflow / Instructions
When asked to create or modify an AGENTS.md file, strictly follow these steps:
- Check for Existing Files: List the root project directory to check if an agent guidance file (like
AGENTS.md,GEMINI.md,CLAUDE.md,.github/copilot-instructions.md,.cursorrules, etc.) already exists. - Analyze File Structure: Print the tree structure of the repository at an appropriate depth (e.g., using
tree -L 2or equivalent python script/command) to gain a comprehensive understanding of the project's content, tech stack, and module organization. Crucially: Because a git repository is constantly changing, do not analyze the file structure too deeply. Focus on the modular structure. - Scan for Instruction Drift: Analyze existing instruction files for redundant, outdated, or conflicting rules ("mudballs"). Recommend a refactoring step if files are overly cluttered or contradictory.
- Formulate a Plan: Based on your findings, formulate a plan to create or update the agent instruction file.
- Single Root File (Default): By default, maintain a single, comprehensive
AGENTS.mdfile in the root directory. Keep the root file minimal (preferably <500 lines) following the Progressive Disclosure principle. - Conditional Progressive Disclosure: If the file exceeds 500 lines or the project is exceptionally complex, split the file (e.g., using nested
AGENTS.mdor linking to sub-docs likedocs/DATABASE.md). - Multi-Framework Support: If the user specifies multiple frameworks (e.g., Copilot, Claude Code, Cursor), implement the Symlink Strategy to use
AGENTS.mdas the Single Source of Truth (SSOT).
- Single Root File (Default): By default, maintain a single, comprehensive
- Execution: Create or modify the file based on your plan and the project context.
More from zenless-lab/skills
python-docstring-expert
Expert methodology for evaluating, formatting, and generating Python docstrings. Use when creating or updating documentation for Python code, determining if a docstring is necessary based on API exposure, or formatting docstrings for modules, classes, and functions. Load this skill when code changes affect existing docstrings to keep them synchronized.
19cloud-init-crafter
Expert assistant for creating, modifying, and debugging cloud-init scripts. Supports multiple formats (YAML, shell, MIME archives), Jinja templating with instance-data, and multi-platform validation.
9readme-crafter
Use this skill when you need to write, refactor, or improve a project's README.md file. Trigger this anytime the user wants to create documentation, project overviews, or profile pages following best practices and modern aesthetics.
8skill-expert
Comprehensive master guide for designing, creating, editing, updating, and refactoring Agent Skills according to the official agentskills.io specification. Use this when you need to build or modify capabilities for an AI agent.
8skill-crafter
Use this skill when creating a new skill or when modifying, updating, refactoring, restructuring, or reviewing an existing skill. Trigger it for framework-specific and framework-agnostic skill work, including SKILL.md design, folder layout, scripts, references, assets, metadata, and description optimization, even if the user only says "make a skill", "improve this skill", or "refactor the skill".
7secret-scanner
Use this skill when instructed to perform a security scan, find leaked secrets, or check for PII in the codebase or git diffs. Trigger this even if the user broadly asks to check for security issues, leaked credentials, passwords, API tokens, emails, or sensitive data before committing.
7