{"id":66411,"plugin_id":"plugin_asdk_app_69fb0ed2dbc4819182d9d1f0acb2a256","kind":"skill","collection_source":"plugin_package","comparison_source":null,"observed_at":"2026-10-28T12:07:23.598Z","digest":"207e0d98a886545f67cac9ba909eb0942b0f0a85162de0cc476b73af254688e9","against":null,"payload":{"description":"Give concrete feedback on a completed Draw Guess sketch when the user asks how to make it clearer, easier to recognize, or better next time. Ground the critique in the uploaded image and recorded result; do not reveal the hidden assigned subject or use this for unrelated artwork.","included_files":[{"relative_path":"agents/openai.yaml","size_in_bytes":257}],"name":"sketch-feedback","skill_md_contents":"---\nname: sketch-feedback\ndescription: Give concrete feedback on a completed Draw Guess sketch when the user asks how to make it clearer, easier to recognize, or better next time. Ground the critique in the uploaded image and recorded result; do not reveal the hidden assigned subject or use this for unrelated artwork.\n---\n\n# Sketch Feedback\n\n## Inputs\n\n- Use the completed round image exposed through model-visible widget state: `modelContent.drawnImageId`, `modelContent.imageUrl`, or `imageIds`.\n- Use the matching `guess-subject` result, if available, and any technique the user wants to improve.\n- Treat an intended subject volunteered after the guess as user-provided context.\n\n## Workflow\n\n1. Match the image and result to the requested round using the visible conversation. If the association is unclear, ask which round to review.\n2. Inspect the actual image. If it is unavailable, explain that visual feedback needs the drawing; guide the user to the saved round's Ask ChatGPT to guess action to refresh its URL. Do not infer image contents from a file ID or result alone.\n3. Complete a pending guess only if the user has explicitly requested guessing for this active round. Choose one exact value from the registered `guess-subject` subject enum and call it once with `guess` and optional `reason`. Otherwise, use the existing result without submitting another guess.\n4. Identify one visible feature that communicates the object effectively, such as silhouette, proportion, or a distinctive detail. If the image is blank or too sparse to support a strength, say so rather than inventing one.\n5. Suggest two specific changes to recognizable features. Tie each change to something visible in the image and explain how it would reduce ambiguity.\n6. If the guess was incorrect and a subject-specific critique needs the intended subject, ask the user what they were drawing. Until they answer, discuss only visible clarity and the recorded guess.\n7. Finish with one technique to prioritize next time. Start another round only when the user asks to play again.\n\n## Output\n\nGive a short critique with one supported strength, up to two actionable changes, and one next-round focus. For a blank image, explain the lack of visual evidence and give one first step instead. Keep the tone constructive and scale the advice to a quick sketch made within the game's 60 seconds.\n\n## Boundaries\n\n- Treat `is_correct` as the recorded outcome, not as a measure of drawing quality. Describe possible causes of confusion as interpretations, not proven explanations.\n- Keep the assigned subject hidden. Do not access `privateContent`, decrypt metadata, or probe `guess-subject` with alternatives to discover the answer.\n- Do not claim to edit the drawing, observe stroke order, measure drawing time, or recover strokes from a PNG.\n- Do not rescore an older image after another `play-round` call; the server stores only the latest subject for the conversation.\n"},"changes":[],"summary":"First saved snapshot. No earlier version is available for comparison.","summary_kind":"deterministic","summary_metadata":{}}