/sg-visual-run — Execute Visual Tests

Execute YAML test manifests using agent-browser (Playwright CLI). Hybrid execution: mechanical steps run directly, complex assertions delegate to LLM evaluation.

Model guidance: This skill runs scripted steps (click, fill, screenshot) plus lightweight LLM assertions. A top-tier reasoning model provides no measurable quality gain here — prefer a fast, capable model to conserve quota.

Invocations

Command	Behavior
`/sg-visual-run`	Interactive — asks what to test
`/sg-visual-run <text>`	Natural language — figures out what tests to run
`/sg-visual-run --from-audit`	Run tests for `impacted_ui_routes` from `visual-tests/_results/audit-results.json`
`/sg-visual-run --from-logic`	Run tests for `impacted_ui_routes` from `visual-tests/_results/logic-results.json`
`/sg-visual-run --from-process`	Run tests for `impacted_ui_routes` from `visual-tests/_results/process-results.json`
`/sg-visual-run --from-audit --from-logic --from-process`	Union mode — merge routes from every selected result file
`/sg-visual-run --diff=main`	Run tests impacted by changes since `main`
`/sg-visual-run --all`	Full suite (skip interactive menu)
`/sg-visual-run --regressions`	Re-run tests that failed last run