Start Here
September 20, 2026 ยท View on GitHub
Install the toolkit, then describe your task with /do. The router selects the relevant agent, skill, and checks.
What You Need
Install Claude Code, then confirm it works:
claude --version
Codex CLI, Factory, and Reasonix are optional. The installer mirrors the components each runtime supports. Use /do in Claude Code, Factory, and Reasonix; use $do in Codex.
If you plan to use an optional runtime, verify it before installation with codex --version, factory --version, or reasonix --version. Claude Code has the broadest hook coverage; the other runtimes receive only the components and lifecycle events they support.
Install
git clone https://github.com/notque/vexjoy-agent.git
cd vexjoy-agent
./install.sh
Choose symlinks if you want git pull updates to appear immediately, or copies for a stable snapshot. The installer puts the toolkit in ~/.claude/ and creates the supported mirrors for other installed runtimes.
Verify the result:
python3 ~/.claude/scripts/install-doctor.py check
python3 ~/.claude/scripts/install-doctor.py inventory
check validates the installation, settings, hook paths, and local data access. inventory shows what each runtime can see. Rerun ./install.sh after pulling changes if you chose copies.
Codex users
Codex integration requires v0.144.1 or later. The current test suite covers 70 hook registrations: 29 run through native Codex events, 30 are adapter-backed, and 11 are unsupported. That is 59 supported registrations in total, but not full Claude Code parity.
After installation or a hook update, run /hooks in Codex and approve changed definitions. Codex does not run changed hook commands until they are trusted.
Try It
Open a project directory and start Claude Code:
claude
Then ask for an overview or give it a concrete task:
/do what can you do?
/do give me an overview of this codebase
/do debug why this test fails
/do write a blog post about [topic]
You do not need to choose an agent or skill. State the outcome you want; /do handles routing and applies the relevant workflow and verification.
If a check reports missing or stale mirrors, rerun ./install.sh, then repeat the two verification commands. The inventory output is the authoritative view of what the installed runtimes can currently discover.
What Was Installed
- Agents provide domain knowledge.
- Skills define reusable methods such as debugging, review, research, and writing.
- Hooks run checks at configured lifecycle events.
- Commands provide entry points such as
/do. - Scripts perform deterministic operations.
These components load when you start Claude Code in any directory. See the root README for architecture, runtime differences, and implementation details.
Where Next
- For Knowledge Workers: writing, research, data analysis, moderation, and HTML artifacts.
- For Developers: architecture, extension points, agents, and skills.
- For AI Power Users: routing, hooks, and pipeline internals.
- For AI Agents: machine-dense component inventory.