Deep Init Skill Creates comprehensive, hierarchical AGENTS.md documentation across the entire codebase. Core Concept AGENTS.md files serve as AI readable documentation that helps agents understand: What each directory contains How components relate to each other Special instructions for working in that area Dependencies and relationships Hierarchical Tagging System Every AGENTS.md (except root) includes a parent reference line: This MUST be visible prose, never an HTML comment. Claude Code strips HTML comments before a memory file reaches the model, so <! Parent: ... , <! Generated: ... and a commented MANUAL boundary are invisible to the agent that is supposed to act on them. This creates a navigable hierarchy: Loading Model Stock Claude Code discovers nested CLAUDE.md by basename and never loads a nested AGENTS.md ; a root @AGENTS.md import does not reach subdirectories either. The nested files generated here are delivered by OMC's PostToolUse directory context injector, which walks up from each accessed file and injects the nearest AGENTS.md / README.md once per session. That means the hierarchy requires OMC's hooks to be active. When OMC hooks are disabled ( DISABLE OMC=1 , OMC SKIP HOOKS=post tool use ) or the plugin is not registered, add a root level CLAUDE.md symlink or @AGENTS.md import for the root file and expect nested files to stay unread. AGENTS.md Template Execution Workflow Step 1: Map Directory Structure Step 2: Create Work Plan Generate todo items for each directory, organized by depth level: Step 3: Generate Level by Level IMPORTANT : Generate parent levels before child levels to ensure parent references are valid. For each directory: 1. Read all files in the directory 2. Analyze purpose and relationships 3. Generate AGENTS.md content 4. Write file with proper parent reference Step 4: Compare and Update (if exists) When AGENTS.md already exists: 1. Read existing content 2. Identify sections : Auto generated sections (can be updated) Manual sections (the Manual Notes heading and everything under it are preserved) 3. Compare : New files added? Files removed? Structure changed? 4. Merge : Update auto generated content Preserve manual annotations Update timestamp Step 5: Validate Hierarchy After generation, run validation checks: Check How to Verify Corrective Action Parent references resolve Read each AGENTS.md, check the Parent context: path exists Fix path or remove orphan No orphaned AGENTS.md Compare AGENTS.md locations to directory structure Delete orphaned files Completeness List all directories, check for AGENTS.md Generate missing files Timestamps current Check the Generated: / Updated: dates Regenerate outdated files Validation script pattern: Smart Delegation Task Agent Directory mapping explore File analysis architect Content generation writer AGENTS.md writes writer Empty Directory Handling When encountering empty or near empty directories: Condition Action No files, no subdirectories Skip do not create AGENTS.md No files, has subdirectories Create minimal AGENTS.md with subdirectory listing only Has only generated files ( .min.js, .map) Skip or minimal AGENTS.md Has only config files Create AGENTS.md describing configuration purpose Example minimal AGENTS.md for directory only containers: Parallelization Rules 1. Same level directories : Process in parallel 2. Different levels : Sequential (parent first) 3. Large directories : Spawn dedicated agent per directory 4. Small directories : Batch multiple into one agent Quality Standards Must Include [ ] Accurate file descriptions [ ] Correct parent references [ ] Subdirectory links [ ] AI agent instructions Must Avoid [ ] Generic boilerplate [ ] Incorrect file names [ ] Broken parent references [ ] Missing important files Example Output Root AGENTS.md Nested AGENTS.md Triggering Update Mode When running on an existing codebase with AGENTS.md files: 1. Detect existing files first 2. Read and parse existing content 3. Analyze current directory state 4. Generate diff between existing and current 5. Apply updates while preserving manual sections Performance Considerations Cache directory listings Don't re scan same directories Batch small directories Process multiple at once Skip unchanged If directory hasn't changed, skip regeneration Parallel writes Multiple agents writing different files simultaneously

deepinit