← ClaudeAtlas

chromeagentlisted

Use when a task needs the user's own running Chrome, with their real profile, cookies, logins, and open tabs, driven through the chrome-devtools MCP server. Verifies or performs one-time setup, then attaches and works.
rivaldofwijaya/chromeagent-skill · ★ 0 · AI & Automation · score 72
Install: claude install-skill rivaldofwijaya/chromeagent-skill
# chromeagent Drive the user's **real** Chrome through the `chrome-devtools-mcp` server. Most sessions are already set up; the whole point of this skill is to reach the user's task without ceremony. ## 1. Fast path: try first, do not probe If `chrome-devtools` MCP tools are present in this session, call `list_pages` immediately. That call only succeeds when config, Chrome version, and the `chrome://inspect` opt-in are all already good, so it **is** the check. On success: say nothing about setup, `select_page` on the tab the user already has open, and get on with the task. Run preflight only when `list_pages` fails or the tools are absent. Do not use it as idle polling: reruns and re-probes after remediation or a user action are expected, but do not repeat it when nothing has changed. ## 2. Preflight For project-scoped targets, the following working-directory rule applies. Both setup scripts resolve paths relative to the current working directory, so run them from the project root; the skill's own directory is never the right place. Invoke the scripts by their absolute path from the project-root shell. If the working directory cannot be the project root, use `--out-dir <dir>` (or `-OutDir <dir>`) as the explicit output override. Preflight scans config paths relative to its own current working directory and has no `--out-dir`/`-OutDir` override. If setup writes elsewhere, run preflight from that output directory; otherwise it reports `NOT_CONFIGURED` even though the conf