Skip to content

Guide · Agent skill

How to use the Deep Codebase Init agent skill (deepinit)

Updated · written with AI from the project's own documentation and checked against our catalogue data

The deepinit skill is a set of instructions for an AI coding agent. When you ask for it, the agent walks through your project folder by folder and writes an AGENTS.md file in each directory describing what that directory contains, how it connects to its neighbours, and how an AI should work there. This guide explains what the skill does, how to install it, how to trigger it, and what to check before you run it on your own repository.

Deep Codebase InitYeachan-Heo/oh-my-claudecode · MIT

Quick start

  1. 1Download the ZIP. Get the deepinit ZIP from the download link on this page.
  2. 2Unzip into a skills folder. Extract it into ~/.claude/skills/ for personal use or into the project's .claude/skills/ folder.
  3. 3Or upload in the Claude apps. Go to Settings → Capabilities → Skills and upload the skill.
  4. 4Confirm OMC hooks are active. Make sure the oh-my-claudecode plugin is registered and DISABLE_OMC and OMC_SKIP_HOOKS are not disabling its hooks.
  5. 5Ask for the documentation. Request AGENTS.md files for the repository or a specific folder in plain language.
  6. 6Review the changes. Check the generated files and the Manual Notes sections before committing.

What the deepinit skill makes the agent do

Agent skills are folders of instructions that an agent such as Claude Code loads when a task matches them. The deepinit skill covers one job: creating hierarchical AGENTS.md documentation across a codebase.

Each AGENTS.md records the directory's purpose, key files, subdirectories, dependencies and special instructions for AI agents. Every file except the root links to its parent folder with a visible line such as Parent context: ../AGENTS.md, so an agent can move from general to specific. When files already exist, the skill re-checks them and updates auto-generated content, while keeping any section titled Manual Notes intact.

Install the deepinit skill

The skill ships as a ZIP file from the download link on this page. You can install it in either of two ways.

For a personal install, unzip the file into ~/.claude/skills/. For a single project, unzip it into the project's .claude/skills/ folder. You can also upload the skill in the Claude apps under Settings → Capabilities → Skills.

  • Personal: ~/.claude/skills/ (available across your projects)
  • Project: .claude/skills/ inside the repository
  • Claude apps: Settings → Capabilities → Skills, then upload

Set up the oh-my-claudecode hooks first

The nested files only work if the oh-my-claudecode (OMC) plugin is registered and its hooks are active. Stock Claude Code does not load nested AGENTS.md files on its own. OMC's post-tool-use directory-context injector reads the nearest AGENTS.md or README.md as the agent touches files in each folder, once per session.

If OMC hooks are disabled with DISABLE_OMC=1 or OMC_SKIP_HOOKS=post-tool-use, or the plugin is not registered, the agent will not read the nested files. In that case, add a root-level CLAUDE.md symlink or an @AGENTS.md import yourself so at least the root file is loaded.

Trigger the skill with a natural request

You do not need a special command. Describe the outcome you want in plain language and mention the repository or folder. The skill's own description is about deep codebase initialization with hierarchical AGENTS.md documentation, so requests that name AGENTS.md files, folder-level context or onboarding an agent to a repository are the most likely to match.

Good requests are specific about scope. For example, ask the agent to create AGENTS.md files for the whole repository, or to refresh the existing ones and keep the Manual Notes sections. Be aware that the skill spawns helper agents (explore, architect and writer) to map and write the files.

  • Document this whole repository with hierarchical AGENTS.md files, level by level.
  • Update the existing AGENTS.md files under src/ and keep each Manual Notes section.
  • Map the project folders and create AGENTS.md only where a directory has real content.

What the run does, step by step

The skill maps the directory tree with an explore agent, excluding folders such as node_modules, .git, dist, build, __pycache__, .venv, coverage, .next and .nuxt. It then plans work by depth level and generates parents before children, so every parent reference points to a file that exists.

After writing, it validates the hierarchy: parent references must resolve, orphaned AGENTS.md files are removed, missing ones are generated and outdated timestamps are regenerated. Empty directories are skipped, while directories with only config files get a short AGENTS.md describing their configuration.

find . -name "AGENTS.md" -type f grep -r "\*\*Parent context:\*\*" --include="AGENTS.md" .

Review and keep your manual notes

Anything you write under the Manual Notes heading is treated as yours and kept on regeneration. Put project rules that the generator cannot infer there. Keep the heading exact, because the skill uses it to find the preserved section.

Because the skill creates and overwrites AGENTS.md files across the tree, review the diff before committing. Commit your work first so you can revert if the output is not what you wanted.

Safety notes: scripts and network

The skill runs scripts, so read the SKILL.md and any bundled scripts before you install it, as you would with any code you run in your repository. It does not use the network, according to the listing.

Expect many new files. The skill is not suited to repositories where you do not want an AI to write AGENTS.md files throughout the tree.

What people use it for

  • Onboarding an AI agent to a large or unfamiliar repository by giving each folder its own context file.
  • Keeping folder-level documentation current in a team that uses coding agents daily.
  • Maintaining an oh-my-claudecode project so nested context loads when agents touch those files.
  • Refreshing existing AGENTS.md files while preserving hand-written Manual Notes.

Frequently asked questions

Is the deepinit skill free?

The source material for this listing does not state pricing or licence terms beyond the MIT licence shown on this page. Check the repository's licence file for the terms that apply to you.

What do I need to run deepinit?

You need the oh-my-claudecode plugin with its hooks active so nested files are loaded, and Claude Code or a compatible agent that can spawn subagents such as explore, architect and writer.

Why doesn't my agent read the nested AGENTS.md files?

Stock Claude Code does not load nested AGENTS.md files. Check that OMC is registered and that DISABLE_OMC or OMC_SKIP_HOOKS is not turning off the post-tool-use hook. If hooks must stay off, add a root CLAUDE.md symlink or @AGENTS.md import for the root file.

Does deepinit need an internet connection?

According to the listing, the skill does not use the network. It does run scripts, so review the bundled files before installing.

How is this different from a single CLAUDE.md or README?

A single root file describes the whole project. Deepinit adds a file per directory, linked to its parent, so an agent can read general context first and specific context where it works. For a small single-file project, one root instructions file is usually enough.

Will deepinit overwrite my existing AGENTS.md files?

It updates auto-generated content in existing files and keeps anything under the Manual Notes heading. It can still overwrite other content, so commit your work before running it and review the diff.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Deepinit suits developers onboarding agents to a large or unfamiliar repository, and teams that want folder-level context kept up to date, especially if they already use the oh-my-claudecode plugin. Skip it for small projects where one root instructions file is enough, or in repositories where you do not want many AI-written files added to the tree.

Download the skill