Update to Planner5D: Plan, Build, Create

{
  "description": "Generate or refine a photorealistic AI image from a native Planner 5D scene screenshot while preserving geometry, materials and visible lighting. Use the editor's quoted screenshot-to-AI workflow.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 381
    }
  ],
  "name": "planner5d-ai-render",
  "skill_md_contents": "---\nname: planner5d-ai-render\ndescription: Generate or refine a photorealistic AI image from a native Planner 5D scene screenshot while preserving geometry, materials and visible lighting. Use the editor's quoted screenshot-to-AI workflow.\n---\n\n# Planner 5D AI rendering\n\nOpen the intended project through the app's navigation, then read `editorGetContext` and `getWebMcpGuide` help for `editorSubmitRender`. The live request schema defines preparation, generation and status recovery.\n\nFor the person's specific furniture, first complete `planner5d-model-import` and verify its native placement, real scale and readiness. For PDF/plan → AI render, complete `planner5d-floorplan-recheck` and the requested furnishing before capture. Reuse the existing project and verify the source before rendering; do not silently substitute a catalog lookalike. If a furnished interior is requested, an empty plan does not fulfill it. An explicit architecture-only/render-only brief preserves that narrower scope. When comparing product/finish variants, preserve verified geometry and a consistent source camera unless the person asks for different views.\n\n1. Choose the view with native 3D preview/camera controls and inspect the actual image. Ensure needed models and materials are loaded. For camera information, request `includeCamera:true` only when the editor is already in 3D.\n2. Prepare the screenshot-based AI request with the actual expected project hash and a stable request ID. Preparation persists native pixels and returns the signed price/quote; it does not submit conventional scene geometry to the server renderer.\n3. Describe the requested appearance using source evidence. Preserve wall/opening locations, furniture proportions, dark material texture, glass/mirror behavior and visible structural features. Describe enabled lighting and its observed positions; do not invent extra fixtures or alter the layout to improve a picture.\n4. Generate only under the user's authorized credit budget, passing the prepared quote and explicit maximum credits required by the live schema. Native permissions and billing decide eligibility and charges.\n5. Follow the existing generation through status using its returned identity. Check completion, saved output, visual fidelity and the accessible image. A retry must recover the previous request before creating another charged generation.\n\nUse native screenshots/previews for source verification. Keep credentials, signed upload/download tokens and one-use login tickets out of user-facing output. Return the verified render and project link, with any visible fidelity differences. An AI image does not certify room measurements or a construction drawing.\n"
}