Update to drawDash - Sketch Challenge!

{
  "description": "Run a voluntary Draw Guess stroke-limit challenge when the user asks to draw with fewer marks, use ten strokes or fewer, or test how simply they can make an object recognizable. Use the normal round and a user-reported stroke count; do not claim automatic counting, enforcement, or access to stroke history.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 284
    }
  ],
  "name": "minimal-strokes-challenge",
  "skill_md_contents": "---\nname: minimal-strokes-challenge\ndescription: Run a voluntary Draw Guess stroke-limit challenge when the user asks to draw with fewer marks, use ten strokes or fewer, or test how simply they can make an object recognizable. Use the normal round and a user-reported stroke count; do not claim automatic counting, enforcement, or access to stroke history.\n---\n\n# Minimal Strokes Challenge\n\n## Inputs\n\n- Use the requested positive whole-number stroke limit; default to ten when unspecified. Clarify an invalid limit before starting.\n- Count one continuous drawing gesture, from contact to lift, as one stroke. Ask the user to count the strokes remaining in their submitted drawing, excluding strokes removed with Undo or Clear.\n- Use the saved image, its confirmed initial `guess-subject` result, and the user's reported count.\n\n## Workflow\n\n1. State the limit, counting rule, and challenge goal: make the assigned object recognizable without exceeding the limit. Explain that the widget does not count or enforce strokes.\n2. Give one brief strategy, such as prioritizing the outline and one distinctive feature. Keep it subject-independent so the assigned word remains hidden from the model.\n3. Start a normal round with `play-round`, using only optional `reason`, when the user is ready. Apply the rule to an existing unfinished round if requested rather than restarting it automatically.\n4. Preserve the 60-second drawing, finish/upload, and explicit Ask ChatGPT to guess flow. Do not request the assigned word or count in advance of the initial visual guess.\n5. Inspect the latest uploaded drawing from model-visible state, then call `guess-subject` once with one exact subject enum value and optional `reason`. Restore image access before guessing if necessary.\n6. Ask for the user's stroke count after the initial guess if it has not been supplied. Accept a nonnegative whole-number count; clarify invalid counts before evaluating compliance. Label the count as user-reported; never estimate it from overlapping marks in the PNG.\n7. Report recognition and limit compliance separately. Mark the challenge met only when the guess was correct and the reported count is within the limit. Mark it unmet when either condition is known to fail; otherwise leave it unconfirmed.\n8. Suggest one image-grounded simplification to try next time. Reuse the limit for another random round only if the user asks to continue; do not promise a repeat of the same subject.\n\n## Output\n\nReport the guess and correctness, the stroke limit, the user-reported count or its absence, the challenge outcome, and one concrete simplification suggestion.\n\n## Boundaries\n\n- Treat count compliance as self-reported. Do not claim stroke telemetry, undo history, automatic disqualification, or tool arguments that set a limit.\n- Do not access private state, decrypt metadata, or probe for the hidden subject with repeated scoring calls.\n- Keep the initial result as the recognition outcome even if the user later explains the drawing.\n- Do not treat failed recognition as proof that a drawing used too few strokes; ground simplification advice in visible features.\n"
}