templates/skills/testing/browser-manual/SKILL.md
Manual browser UI verification via MCP snapshots with reproducible steps and observed behavior; not automation design
npx skillsauth add ozerohax/assistagents testing-browser-manualInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
<input_requirements> <required>Environment URL and access</required> <required>Key user scenarios list</required> <required>Success criteria for each scenario</required> <optional>Supported browsers and versions</optional> <optional>Key resolutions/breakpoints</optional> <optional>Test accounts and data</optional> <optional>Environment constraints (rate limits, feature flags)</optional> </input_requirements>
<preparation> <steps> <step>Verify the environment and build freshness</step> <step>Create a new context/page via MCP</step> <step>Do the first navigation to start network/console logging</step> <step>Fix viewport, language, and locale</step> <step>Clear cookies/localStorage when needed</step> </steps> </preparation><execution_rules> <rule importance="critical">Steps are executed via agent tools and are reproducible</rule> <rule importance="critical">Actions are done by snapshot ref, not by screenshot</rule> <rule importance="critical">Verify UI and system reactions (errors, states)</rule> <rule importance="high">Cover positive and negative scenarios</rule> <rule importance="high">Verify accessibility (keyboard navigation, focus, aria)</rule> <rule importance="high">Verify responsive behavior at key breakpoints</rule> <rule importance="high">Note dependencies on data and state</rule> <rule importance="medium">Record regression signals and visual artifacts</rule> </execution_rules>
<coverage> <focus> <item>Primary user paths</item> <item>Form validation and errors</item> <item>Loading states and empty states</item> <item>Navigation and routing</item> </focus> </coverage><quality_rules> <rule importance="critical">Expected outcome is stated unambiguously</rule> <rule importance="high">No duplicate scenarios with different wording</rule> <rule importance="high">UI errors and network errors are distinguished and verified separately</rule> <rule importance="high">Network/console logs are checked after navigation or actions</rule> <rule importance="high">Viewport/browser are recorded in the result</rule> <rule importance="medium">If result recording is needed, use a single consistent format</rule> </quality_rules>
<do_not> <item importance="critical">Do not test production without permission</item> <item importance="high">Do not use real user data</item> <item importance="high">Do not ignore errors in console and network logs</item> <item importance="high">Do not rely on visual match without checking DOM state</item> </do_not>
<agent_checklist> <item importance="critical">Test goal and success criteria are stated</item> <item importance="critical">Environment and access are confirmed</item> <item importance="high">Start route and session state are defined</item> <item importance="high">Viewport/browser/locale are fixed</item> <item importance="high">Steps and expected UI states are listed</item> <item importance="high">Network/console checks are planned</item> <item importance="medium">Test boundaries and exclusions are captured</item> </agent_checklist>
<example_checks> <check>Verify validation error display for an empty required field</check> <check>Verify correct handling of 401/403 for an expired session</check> <check>Verify keyboard form navigation and visible focus</check> </example_checks>
testing
Visual regression (baseline diffs, UI stabilization)
testing
Design detailed reproducible test cases (Given/When/Then, one check per case); not smoke/regression checklists
development
Mobile testing (native/hybrid/web, gestures, offline, interruptions)
testing
Load testing (k6/JMeter scenarios, benchmarks)