browser-check

Run the skill's check.py script with an explicit --url like http://127.0.0.1:5173/ and a fresh --output directory. It launches a sandboxed headless Chrome, captures initial and final PNG screenshots plus a dom.json observation file, and writes a report.json with status and hashes.

Provide a reviewed JSON actions file with up to 16 actions such as click, fill, press, scroll, wait-for, and screenshot. Click and fill require exactly one CSS selector match, and password, file, hidden, and one-time-code fields are refused.

No. Only explicit http://127.0.0.1 URLs with unprivileged ports are accepted, and browser-level rules restrict requests to that single local origin. Public websites, authenticated browsing, CDP attachment to other browsers, and cloud browser agents are explicitly out of scope.

No. Exit code 0 only means the requested capture and actions completed with clean owned-process cleanup. Screenshots are not performance benchmarks, complete interaction tests, or accessibility audits; page errors and untested behavior must be reported separately.

Exit 1 indicates refusal or incomplete operation, such as a missing verified runtime, sandbox checks failing, unexpected dialogs or downloads, or cleanup problems. Read report.json errors and retained evidence rather than treating the run as clean.

Run a second bounded capture with --width 390 --height 844 and --reduced-motion. The script emulates prefers-reduced-motion and the narrow viewport, producing separate screenshots and DOM observations for comparison.