GRI AI / Developer release

Claude Code subagents

Supercharge Claude Code. The right agent. The right settings. A second look before code runs—powered by the Liz AI Platform's auto-classifier.

What it does

Two tools, one goal: better output.

A shared set of Claude Code subagents, each pinned to the right model and effort level. The agents are language- and framework-agnostic: they read each project's own conventions, build files and test commands instead of assuming them. Choose routing or risk-check mode at installation; these modes currently run separately.

Route with precision

A prompt router matches each request to the right specialist agent, with model and effort settings tuned to the task.

Keep auto mode in check

A risk check reviews every file edit and shell command before it runs—even in auto mode. High-risk actions pause for your confirmation.

Get started

Install in one line.

The installer adds the agents and the risk check. Your browser opens to sign in to status402; then restart Claude Code. It needs Claude Code, git, python3 and a status402 account.

Install (risk check, default)

Prefer the prompt router? Set CLAUDE_TOOLS=router. Use none to install the agents only. Re-run the installer to update the agents or switch tools.

Install with router

To uninstall, run:

Uninstall

Transparency

What gets sent where.

The agents are local files. The two optional tools call the status402 service, which you sign in to through your browser.

Risk check (the default)

Sends the tool's name, the command or file path, and the text being written—before every shell command and file write or edit.

Router

Sends your prompt, and the agents' names and descriptions—on every prompt.

Agents only

With CLAUDE_TOOLS=none, nothing is sent.