Prove Your Agent's Work with Evidence-Driven Testing

Where this fits

You've just finished a UI change and your agent claims it works, but the PR reviewer wants more than prose. Instead of scrambling to paste screenshots or write a separate test report, your agent can now record its own hands-on testing session—complete with structured annotations for each assertion—and attach the video plus a results summary directly to the PR and the tracker issue. This skill turns "trust me" into "watch me."

Why agents benefit

• Captures visual evidence during computer-use testing, so the agent can demonstrate actual behavior rather than describe it.
• Works in headless environments via scripted screenshots and probes, so CI or remote agents still produce proof.
• Posts the video and summary automatically to GitHub PRs and issues using gh, closing the loop with stakeholders.
• Annotates each test step with assertion results, making failures and passes explicit in the recording.
• Handles non-UI changes too—measured numbers, output pairs—so evidence isn't limited to pixels.

Practical scenarios

Verifying a responsive layout change

Your agent opens the app in a browser, resizes the viewport across breakpoints, and records each state. Annotations show that the nav collapses correctly and content doesn't overflow. The video is attached to the PR, and the reviewer sees exactly what changed.

Testing a multi-step form with validation

The agent navigates through a checkout flow, submits invalid inputs, confirms error messages appear, then completes a successful order. Each step is recorded with an assertion badge (pass/fail). The final video is posted to the issue, giving QA a replayable trace.

Validating a backend API response (non-UI)

Even without a GUI, the agent runs scripted probes against an endpoint, measures response times, and captures output pairs (request/response). The recorded log and summary numbers become the evidence, attached to the PR for performance review.

Add it to your agent workflow

Install the skill with:

npx skills add michaelshimeles/skills --skill evidence-driven-testing

Then, in your agent's task, instruct it to use the skill when a change needs verifiable proof. For example:

Test the new login flow via computer use. Record the session, annotate each assertion, and post the video and summary to PR #42 and issue #17.

The skill handles the recording, annotation, and posting—provided gh is authenticated and the app is running.

Compare before adopting

Before wiring this in, compare it with your current testing workflow. Look at quality signals like whether the recorded video is actually watchable and annotated clearly. Check maintenance freshness—is the skill actively updated? Consider alternatives like simple screenshot tools or manual video capture, and weigh the setup cost (GUI or headless) against the benefit of automated evidence. Also confirm it fits your agent's existing computer-use capabilities.

Why it is worth tracking

This skill addresses a real pain point: the gap between what an agent claims and what a reviewer can verify. With 332 stars and active maintenance, it's gaining traction in the agent-skills ecosystem. The ability to produce annotated, postable evidence in both GUI and headless contexts makes it a versatile addition. Evaluate it when your agents start making user-facing changes that require sign-off—it could become the default way your team trusts agent output. For a closer look at the skill's metadata and version history, see the OpenAgentSkill listing.