Send a crowd of AI virtual users at a URL and return the damage summary.

Only test sites the user owns or has permission to test. A run drives real browsers and typically takes several minutes; more personas and higher verify levels take longer.

Args: url: The website to test. personas: Comma-separated built-in persona names, "all", or "none". mob: Additionally generate this many random virtual users. seed: Random seed making a mob reproducible. goal: Extra mission given to every persona. verify: Verification depth: "none", "detective" (independent agent re-checks critical/major findings), "cross" (adds a Playwright cross-engine probe; default), or "tribunal" (adds a Microsoft Webwright agent as a third harness). provider: "auto" (default, from env keys), "anthropic", or "openai". model: Model id override. max_steps: Max browser actions per persona. concurrency: How many personas browse at once. out_dir: Report directory (default: ~/crowd-test-reports/-).

List the built-in ringleader personas and what each one catches.

Preview the random mob members a given count/seed would generate.

Return the newest crowd-test markdown report under a directory.

Defaults to ~/crowd-test-reports, where run_crowd_test writes.