Evidence for agent runs
Give agent reports a screenshot you can inspect
A coding or QA agent can capture a deployed page from its shell, then place the returned asset URL in its report. Humans can inspect the rendered result instead of trusting a summary based only on source code.
export SCRNIFY_API_KEY=...
evidence_url="$(scrnify capture "$DEPLOY_URL" \
--type image --format png \
--width 1440 --height 900)"
printf 'Visual evidence: %s
' "$evidence_url"Report output
One URL identifies the Capture. It is evidence of rendered page state, not proof that every requirement passed.
Run sheet
Define what the image is meant to prove
- 01
State the expected view
Tell the agent which deployed URL, viewport, and page region matter. Use a CSS selector when one stable component—not the whole page—is the subject of the task.
- 02
Load the CLI contract
scrnify skill prints machine-readable command help, environment variables, examples, and exit behavior. The agent can inspect that contract before constructing a command.
- 03
Capture after deployment
Run against a public, reproducible URL after code is deployed. The hosted browser does not see an agent’s local dev server or inherit its already-open browser tab.
- 04
Report limits with the URL
Record viewport and target URL beside the evidence. Say whether anyone inspected the image; capture success alone does not evaluate spacing, copy, accessibility, or acceptance criteria.
Before sharing
Make evidence auditable
- Include target URL and viewport dimensions in the report so another person can reproduce the Capture.
- Use stable deployed states instead of pages whose content changes per session or request.
- Keep API credentials in the agent runtime environment, outside prompts and final reports.
- Use a selector only when it is stable and unique; an invalid selector makes the remote Capture fail.