Update to drawDash - Sketch Challenge!

{
  "description": "Explain a recorded Draw Guess guess or confusing result when the user asks why ChatGPT chose that subject, what visual cues supported it, or why their sketch was misrecognized. Reinspect the matching uploaded image and use the existing result; do not submit another guess or claim access to the hidden answer or unrecorded reasoning.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 256
    }
  ],
  "name": "explain-drawing-guess",
  "skill_md_contents": "---\nname: explain-drawing-guess\ndescription: Explain a recorded Draw Guess guess or confusing result when the user asks why ChatGPT chose that subject, what visual cues supported it, or why their sketch was misrecognized. Reinspect the matching uploaded image and use the existing result; do not submit another guess or claim access to the hidden answer or unrecorded reasoning.\n---\n\n# Explain Drawing Guess\n\n## Inputs\n\n- Use the requested round's recorded `guess` and `is_correct` from `guess-subject`.\n- Use its uploaded image from model-visible widget state or an image already visible in the conversation.\n- Use an intended subject supplied by the user only as comparison context.\n\n## Workflow\n\n1. Identify the recorded guess and match it to the requested image. If either is missing or belongs to an ambiguous round, ask for the missing context instead of inventing an explanation.\n2. Inspect the image itself, using the visible `modelContent.imageUrl`, `drawnImageId`, or `imageIds` as available. If inspection fails, say that the result alone cannot establish which visual cues influenced recognition; guide the user to refresh access through the round widget.\n3. Describe up to two visible features that resemble the recorded guess, such as the outline, relative size of parts, or a characteristic detail. Explain how those features could support the choice without claiming to reconstruct unrecorded reasoning.\n4. Use `is_correct` to state whether that guess matched the assigned subject. If it was incorrect, identify a visual ambiguity when the image supports one.\n5. If the user supplied the intended subject, compare one distinguishing feature with the recorded guess. Otherwise, do not name an actual answer; invite the user to share it only if they want that comparison.\n\n## Output\n\nGive a concise explanation containing the recorded outcome, the visible supporting cues, and any important uncertainty. Keep a full drawing critique separate; use `sketch-feedback` when available if the user asks how to improve.\n\n## Boundaries\n\n- Use an existing result only. Do not call `guess-subject` to rescore the image, try alternatives, or reveal the hidden subject.\n- Do not start a new round merely to explain a previous result.\n- Do not access private widget content, decrypt metadata, or infer an assigned subject from an incorrect result.\n- Describe ambiguous features as possible explanations, not proof that the drawing or recognition process failed for a particular reason.\n"
}