Browser debug

Command

Give an agent a token-efficient, scriptable way to debug a web page — screenshots, accessibility snapshots, console and network logs, Lighthouse — without bloating its context.

cckit debug drives chrome-devtools-axi (Kun Chen), which runs a self-contained Chrome DevTools bridge and emits TOON-encoded output. TOON is a compact text format, so results stay small in an agent’s context.

Built on chrome-devtools-axi

The browser layer is chrome-devtools-axi by Kun Chen — a self-contained Chrome bridge for agents. cckit only wraps and auto-detects it; all credit for the browser tooling goes there. If it helps you, ⭐ the repo.

New here? Why this exists

Debugging a web page normally means a person clicking around a browser. An agent cannot do that. cckit debug exposes the same browser tools as plain commands an agent can run and read, so the agent can see what is on the page.

Optional + auto-detected

cckit is pure bash; axi needs Node and Chrome. So this is not a hard dependency — cckit detects it at call time and degrades cleanly:

• Node/npx and Chrome present → cckit debug drives axi.
• Otherwise → it prints how to enable the capability and exits 0. An unattended run is never hard-failed by a missing optional tool.

Check availability without running anything:

cckit debug --check

Enabling it

npm i -g chrome-devtools-axi      # or rely on `npx -y chrome-devtools-axi` (Node present)

Override the command cckit invokes with an environment variable — useful for a pinned version or a wrapper:

export CCKIT_AXI=<command>

Usage

Everything after cckit debug is forwarded verbatim to axi, so cckit never has to track axi’s evolving subcommand surface:

cckit debug <axi-args...>         # passed straight through — see the axi docs for subcommands