| name | agent-browser |
| description | Debug browser automation flows (auth, redirects, SPAs) using agent-browser CLI. Use when diagnosing Playwright scripts, SSO flows, or any browser-based issue where you need to see actual page state. |
| user-invocable | true |
| allowed-tools | Bash, Read, Glob, Grep |
| argument-hint | ["url-or-description"] |
Use agent-browser (installed via brew) to visually inspect browser flows.
Workflow
- If a process generates the URL (e.g.
aws sso login), start it with BROWSER=/usr/bin/true to suppress its browser, capture the URL from stdout
- Open the URL:
agent-browser open <url> (headless) or agent-browser --headed open <url> (if user needs to interact, e.g. enter credentials)
- At each step:
agent-browser screenshot /tmp/step-N.png and read it, agent-browser snapshot -i for interactive element refs, agent-browser get url for current URL
- Interact via
agent-browser click @ref / agent-browser fill @ref "text" or let user interact in headed mode
- Repeat screenshot/snapshot after each transition until flow completes
agent-browser close when done
Authenticated flows
Use --session-name to persist auth (cookies/storage) across runs:
- First run (user authenticates):
agent-browser --headed --session-name <name> open <url>
- Subsequent runs reuse the session headlessly
State saved to ~/.agent-browser/sessions/. See agent-browser state --help for manual save/load/cleanup.
Run agent-browser --help for full command reference.
