agd:init-agents-md

Create a compact, lean AGENTS.md file tailored to the current project. Default to a single file; ask about references folder only if the project is complex.

Load Context

Load both skills for comprehensive context:

1. AGENTS.md format - Load agents-md skill:

   Use Skill tool with skill="agd:agents-md"
   

   Provides: sections, structure, best practices

2. Commit format - Load conventional-commits skill (for PR/Commit Guidelines section):

   Use Skill tool with skill="agd:conventional-commits"
   

   Provides: Conventional Commits types, style rules, examples

Process

1. Analyze the project:

   • Check for package.json, Cargo.toml, go.mod, pyproject.toml, etc.
   • Identify test commands from package scripts or config files
   • Detect linting/formatting tools (eslint, prettier, rustfmt, black)
   • Find build commands and development scripts

2. Detect project type:

   ls -la
   cat package.json 2>/dev/null | head -50
   

3. Gather context:

   • README.md for project description
   • Existing documentation
   • CI/CD configuration (.github/workflows, etc.)

4. Generate compact AGENTS.md using the format from the skill:

   • Apply recommended sections from skill knowledge
   • Fill in detected project-specific content
   • Keep content concise and actionable (target: 50-80 lines)
   • Focus on essential commands and conventions only
   • Use relative paths for references (e.g., ./docs/agents/testing.md, ../api/AGENTS.md)
   • For monorepos: reference sibling packages with relative paths (e.g., ../shared/utils/)

5. Write the file to the specified path (default: ./AGENTS.md)

6. Validate the generated file:

   • Check all commands are executable
   • Ensure sections are complete and relevant
   • Confirm file is compact (under 100 lines)
   • Verify no placeholder content

7. Ask about references folder using AskUserQuestion: Ask: "Would you like me to create a references folder for detailed documentation?"

   Options:

   • "No, single file is enough" (default) - Simple AGENTS.md satisfies
   • "Yes, create docs/agents/" - Create references folder with detailed docs
   • "Yes, use custom path" - Ask for custom path

   Only ask if:

   • Project has complex architecture
   • Many project-specific patterns to document
   • Generated AGENTS.md exceeds 80 lines

8. Ask about symlinks (from skill best practices):

   Call AskUserQuestion:

   {
     "questions": [{
       "question": "Would you like me to create symlinks for other AI agents?",
       "header": "Symlinks",
       "options": [
         {"label": "No symlinks", "description": "Keep AGENTS.md as the only file."},
         {"label": "Create CLAUDE.md", "description": "Symlink CLAUDE.md → AGENTS.md for Claude Code compatibility."},
         {"label": "Create all symlinks", "description": "Create CLAUDE.md, .cursorrules, and .windsurfrules all pointing to AGENTS.md."}
       ]
     }]
   }
   

   • "No symlinks" → Done
   • "Create CLAUDE.md" → Run ln -sf AGENTS.md CLAUDE.md
   • "Create all symlinks" → Run ln -sf AGENTS.md CLAUDE.md && ln -sf AGENTS.md .cursorrules && ln -sf AGENTS.md .windsurfrules

Template Detection

AskUserQuestion Format

Use this format when asking about references folder:

questions: [
  {
    "question": "Would you like me to create a references folder for detailed documentation?",
    "header": "References",
    "options": [
      {
        "label": "No, single file is enough",
        "description": "Keep AGENTS.md as a simple, compact file. Recommended for most projects."
      },
      {
        "label": "Yes, create docs/agents/",
        "description": "Create references folder with detailed docs for architecture, patterns, tutorials."
      },
      {
        "label": "Yes, use custom path",
        "description": "Specify a custom path for the references folder."
      }
    ]
  }
]

Notes

• Default to simple: Always start with a compact single-file AGENTS.md
• Do not overwrite existing AGENTS.md without confirmation
• Keep generated content concise and actionable
• Include actual commands the agent can execute
• Follow the format and best practices from the loaded skill
• Only create references folder when user explicitly wants it
• Validate generated file before completing
• Always load the agents-md skill first for context