name: bad
description: Browser Agent Driver CLI operator for browser automation, UI/design audits, auth state, showcases, and benchmark runs. Triggers: "run bad", "browser agent", "design audit", "webbench", "automate this site".
bad
Use the bad CLI for real browser work: automation, evidence capture, design audits, site rips, auth-state setup, showcases, and browser benchmarks.
Do not fake browser results; every claim must come from a run artifact, screenshot, DOM extract, trace, or saved report.
Start
- Verify the binary:
bad --version.
- If missing, install
@tangle-network/browser-agent-driver or run it from source; stop if neither is available.
- Run
bad --help or the specific subcommand help before relying on flags.
- Prefer a narrow first run over a broad audit; expand only after the command and target are proven.
Commands
bad run executes a browser task from a goal, URL, or case file.bad design-audit scores product/visual/trust issues, extracts tokens, compares sites, or rips a page.bad showcase captures hero shots, full-page screenshots, walkthroughs, dark mode, and retina assets.bad auth captures, validates, or injects authenticated browser state.bad runs lists historical runs and helps resume or fork prior work.
See references/full-reference.md for the full CLI command catalog, auth patterns, config shape, and examples.
Operating Rules
- For UI or design claims, inspect actual screenshots or extracted DOM/style data.
- For authenticated flows, validate the auth state before running the test or benchmark.
- For site ripping or clone prep, use
bad design-audit --rip or the current equivalent from bad design-audit --help.
- For benchmark work, record provider, model, mode, profile, case count, pass/fail counts, wall time, and artifact path.
- For patch validation, re-audit after the patch and compare against the prior run.
Then consider
ui-test when the task is specifically QA of a product UI or PR.product-design-audit when the user wants redesign decisions, not only browser evidence.
