Claude Code is powerful on its own. This toolkit makes it much better -- specialized agents for every domain, workflow skills that enforce methodology, and hooks that automate the boring parts. You install it once and it works everywhere.
One thing: Claude Code installed and working. Open a terminal and run:
claude --versionIf that prints a version number, you're good. If not, install Claude Code first and come back.
Optional: if you also use Codex CLI, run codex --version. The toolkit mirrors its skills into ~/.codex/skills and its agents into ~/.codex/agents, so Codex sessions can Read the same domain expertise Claude Code dispatches. Claude Code is still the full runtime for hooks, commands, and scripts.
Command entry points:
- Claude Code:
/do - Codex:
$do
Three commands. Copy-paste them.
git clone https://github.com/notque/claude-code-toolkit.gitcd claude-code-toolkit./install.shThe installer asks one question -- symlink or copy -- then sets everything up. Pick symlink if you want updates via git pull, copy if you want a stable snapshot. Either works fine.
What just happened: the installer linked agents, skills, hooks, commands, and scripts into ~/.claude/, which is where Claude Code looks for extensions. It also mirrored skills into ~/.codex/skills and agents into ~/.codex/agents for Codex, and configured hooks in your settings so they activate automatically.
Run:
python3 ~/.claude/scripts/install-doctor.py check
python3 ~/.claude/scripts/install-doctor.py inventorycheck verifies the install layout, settings, hook paths, learning DB access, and both Codex skill and agent mirrors. inventory shows what Claude and Codex can currently see. If you pull new toolkit changes later and want Codex to pick up new skills or agents, rerun ./install.sh.
Open any project folder. Start Claude Code.
claudeNow try these.
/do what can you do?
The router command is the front door. Use /do in Claude Code and $do in Codex. It reads your request, picks the right agent and skill, and runs it. This one shows you the full routing system -- every domain it handles, every workflow it knows.
/do give me an overview of this codebase
Try this in any project you're working on. It'll read the code structure, identify patterns, and explain what the project does. Works with Go, Python, TypeScript, Kubernetes configs, whatever's in the repo.
/do write a blog post about [topic you care about]
This kicks off a multi-phase pipeline -- research, outline, draft, voice validation. Replace the topic with something real. The output lands in a file you can edit.
/do debug why [describe the problem]
Routes to a systematic debugging skill. It'll gather evidence before guessing, which is the whole point.
Five kinds of things got copied to ~/.claude/:
- Agents -- domain experts for Go, Python, Kubernetes, data engineering, content, and more
- Skills -- reusable workflows like TDD, debugging, code review, article writing, research pipelines
- Hooks -- automation that fires on session start, after errors, before context compression
- Commands -- slash command definitions that wire up user-facing entry points like
/do - Scripts -- Python utilities the agents call for deterministic operations
These load automatically when you start Claude Code in any directory. You don't need to configure anything else.
Depends on what you're here for.
For Developers -- Architecture, extension points, how to build your own agents and skills.
For Knowledge Workers -- Content pipelines, research workflows, moderation, data analysis. No code required.
For AI Power Users -- Routing internals, hook lifecycle, pipeline architecture. The deep stuff.
For AI Agents -- Machine-dense component inventory. If you're an LLM operating in this repo, start there instead.