Example request
Task for agent: 1. Capture current app screens. 2. Draft store-screenshots.json using /openapi.json. 3. POST /v1/renders with a test key. 4. Read lint[]. If status=fail, revise the request and render again. 5. Present final image URLs and billing/upload actions to a human.
Implementation checklist
- Do not invent app UI; use real captures.
- Prefer test keys for drafts.
- Use lint details as revision instructions.
- Stop for human approval before payment or store upload.
Agent-safe boundaries
The optional AI surface can draft copy and layout. The deterministic renderer composites the real screenshots. That separation keeps generated assets auditable.