playwright-cli

Browser Automation with playwright-cli

Mingyi Atlas pentest runner usage

When running in Mingyi Atlas pentest mode, prefer the run_browser_cli tool instead of local Bash. It executes playwright-cli inside a task-scoped browser container from the configured containers/browser-runner image.

Tool input rules:

• Set sessionId to a task-scoped id and reuse it for all browser steps in that task.
• Put only the playwright-cli arguments in argv; do not include playwright-cli itself. Prefer string[]; JSON-encoded string arrays are accepted only for compatibility.
• Include every URL/host touched by the command in targets so scope validation can run.
• Use scopeHosts only when the authorized scope has not already been recorded.
• Use expectedOutputs for files written under /out.
• For screenshot, snapshot, and pdf, run_browser_cli will add a missing /out/... filename automatically so evidence is collected.
• Use action: "close" when the task ends to destroy the browser container.
• Browser runner containers are labeled for stale cleanup, but explicit action: "close" remains the preferred cleanup path.

Example:

{
  "sessionId": "recon-example",
  "argv": ["open", "https://example.com"],
  "targets": ["https://example.com"],
  "purpose": "Open scoped target in browser runner"
}

Follow-up step in the same task:

{
  "sessionId": "recon-example",
  "argv": ["snapshot"],
  "targets": ["https://example.com"],
  "purpose": "Capture scoped page snapshot"
}

Task cleanup:

{
  "sessionId": "recon-example",
  "action": "close",
  "purpose": "Close browser runner for completed task"
}

Quick start

# open new browser
playwright-cli open
# navigate to a page
playwright-cli goto https://playwright.dev
# interact with the page using refs from the snapshot
playwright-cli click e15
playwright-cli type "page.click"
playwright-cli press Enter
# take a screenshot (rarely used, as snapshot is more common)
playwright-cli screenshot
# close the browser
playwright-cli close