cwnu-open-source-community/universal-code/run-verify

Run verification

Use this workflow when a change needs evidence from the running program. Identify the source state, working directory, available launch recipe, required configuration, and expected observable behavior. Inspect repository scripts and offered runtime capabilities. Use an existing suitable task-owned process or start one through available lifecycle tools; do not guess executable paths or take over unrelated listeners.

Choose the smallest check that exercises the requested path. For a CLI or service, inspect exit status, output, and the relevant state change. For a UI, open the page built from the changed source, perform the important user actions, and inspect results and console errors. Cover narrow layouts, focus, error states, or persisted state when the change affects them. Frontend visual work also needs rendered-image inspection when available; DOM checks alone do not prove appearance. Motion needs observations over time, not a single still frame.

Distinguish a build, a mock, a source check, a captured screenshot, a visually inspected screenshot, and an exercised interaction. Report each only as what it is. When an expected result fails, preserve the first useful error and inspect its cause before changing code or retrying. Repeat a check after a relevant repair or changed prerequisite, not merely to obtain a pass. If the current tools cannot launch, browse, or view images, state the specific gap and use available checks without inventing execution.

Keep enough evidence to reproduce the result: source revision or changed-file hashes, command or structured operation, route, relevant viewport and state, observed result, and artifact paths. For the optional ui-evidence.v1 operation, follow its offered schema and source/attempt bindings; its admission and evidence rules apply to that operation. Do not impose that protocol on unrelated browser or command tools.

Clean up only processes and temporary resources owned by this verification, unless the user needs a preview left running. Explain any remaining preview and how it was started. Deliver a concise account of what passed, failed, or was not run. This guidance grants no execution, browser, network, sandbox, or approval authority; use the current Run's offered tools and scope.