Update to 1 Billion Brain Cells

{
  "description": "Review one completed Brain Training game and turn its score and success/failure counts into one practical adjustment and a replay or switch recommendation. Use for single-round analysis and game-completion follow-ups, not comparisons across attempts, live-board hints, or clinical assessment.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 261
    }
  ],
  "name": "brain-training-result-review",
  "skill_md_contents": "---\nname: brain-training-result-review\ndescription: Review one completed Brain Training game and turn its score and success/failure counts into one practical adjustment and a replay or switch recommendation. Use for single-round analysis and game-completion follow-ups, not comparisons across attempts, live-board hints, or clinical assessment.\n---\n\n# Brain Training Result Review\n\n## Inputs and workflow\n\n1. Identify the game from its associated tool call or the user's statement. Read\n   the completed result from model-visible widget `modelContent`, a completed\n   tool output, the completion follow-up, or counts supplied by the user.\n   A launch response with `result: null` is pending, not a zero-score game.\n2. Use the available counts and score. Ask for the game or missing counts only\n   when needed for the requested analysis; give a score-only explanation when\n   that is all the evidence supports. Keep missing values unknown rather than\n   substituting zero.\n3. Interpret scoring using the following game-specific rules. Prefer the\n   reported final score; flag a conflict with the counts instead of silently\n   changing either value. Consult the available tool contract if it differs.\n\n   | Game tool | Success / failure fields | Score |\n   | --- | --- | --- |\n   | `color-switch` | `successfulPrompts` / `failedPrompts` | `(successes - failures) * 100` |\n   | `number-sort` | `successfulRounds` / `failedRounds` | `(successes - failures) * 100` |\n   | `odd-one-out` | `successfulRounds` / `failedRounds` | `successes * 100`; failures do not subtract points |\n   | `path-trace` | `successfulRounds` / `failedRounds` | `(successes - failures) * 100` |\n   | `pattern-recall` | `successfulRounds` / `failedRounds` | `(successes - failures) * 100` |\n   | `word-run` | `successfulWords` / `unsuccessfulWords` | Each accepted word earns `(length - 2) * 100`; each rejected submission costs 100 |\n\n4. When both counts are available, calculate the success share of recorded\n   prompts, rounds, or submissions as `successes / (successes + failures)`.\n   Report it as unavailable if the denominator is zero. Color Switch failures\n   include expired prompts, so do not call this a wrong-answer rate. Do not\n   reconstruct Word Run's score from counts alone; word lengths are not saved.\n5. Explain one observable feature of the result and suggest one adjustment\n   grounded in that game's rules. Treat explanations such as rushing or losing\n   the sequence as possibilities unless the user reported them. Describe\n   performance in this game without inferring IQ, clinical status, or population\n   rank from the score.\n6. Offer a same-game replay with that adjustment or one alternative suited to\n   the user's goal. Launch a fresh game only when the user has requested further\n   play; call its exact registered tool with at most an optional string `reason`.\n\n## Deliverable and boundaries\n\nReturn the score and counts that are known, one short interpretation, one\nactionable adjustment, and a concrete next-game suggestion. Scale the detail to\nthe user's question.\n\nUse only model-visible data or information supplied by the user. The app exposes\naggregate results, not individual answers, reaction times, boards, hidden paths,\nor submitted-word history. Do not claim to have inspected those details or read\nprivate widget metadata. Do not launch a game to retrieve a past result or retry\nan active game merely to obtain data.\n"
}