dogfood

Provide a target URL, a testing scope, and an optional output directory. The agent follows a 5-phase workflow: plan a sitemap, explore pages with browser tools, collect screenshot evidence, categorize issues by severity, and generate a structured report.

The workflow requires browser_navigate, browser_snapshot, browser_click, browser_type, browser_vision, browser_console, browser_scroll, browser_back, and browser_press. These cover navigation, DOM inspection, interaction, screenshots, and console error detection.

Each issue is assigned a severity (Critical, High, Medium, Low) and a category (Functional, Visual, Accessibility, Console, UX, Content) based on the taxonomy in references/issue-taxonomy.md. Issues are de-duplicated and sorted by severity in the final report.

Yes. The workflow checks browser_console after every navigation and significant interaction, specifically to catch silent JS errors, uncaught exceptions, failed network requests, and deprecation warnings that users may not see.

It depends on the availability of the browser toolset and cannot execute areas blocked by authentication or other blockers, which are noted in the report. It performs exploratory testing rather than scripted regression suites or load testing.