<input_requirements> Environment URL and access Key user scenarios list Success criteria for each scenario Supported browsers and versions Key resolutions/breakpoints Test accounts and data Environment constraints (rate limits, feature flags) </input_requirements>

<execution_rules> Steps are executed via agent tools and are reproducible Actions are done by snapshot ref, not by screenshot Verify UI and system reactions (errors, states) Cover positive and negative scenarios Verify accessibility (keyboard navigation, focus, aria) Verify responsive behavior at key breakpoints Note dependencies on data and state Record regression signals and visual artifacts </execution_rules>

<quality_rules> Expected outcome is stated unambiguously No duplicate scenarios with different wording UI errors and network errors are distinguished and verified separately Network/console logs are checked after navigation or actions Viewport/browser are recorded in the result If result recording is needed, use a single consistent format </quality_rules>

<do_not> Do not test production without permission Do not use real user data Do not ignore errors in console and network logs Do not rely on visual match without checking DOM state </do_not>

<agent_checklist> Test goal and success criteria are stated Environment and access are confirmed Start route and session state are defined Viewport/browser/locale are fixed Steps and expected UI states are listed Network/console checks are planned Test boundaries and exclusions are captured </agent_checklist>

<example_checks> Verify validation error display for an empty required field Verify correct handling of 401/403 for an expired session Verify keyboard form navigation and visible focus </example_checks>