Update to Caveman Mode

{
  "description": "Practise a target language through short, alternating dialogue with a supported Persona app character, selective corrections, and a focused retry. Use for character-themed language roleplay at a stated proficiency level. Do not use for translation alone, general concept teaching, pronunciation scoring, or audio practice.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 274
    }
  ],
  "name": "persona-language-roleplay",
  "skill_md_contents": "---\nname: persona-language-roleplay\ndescription: Practise a target language through short, alternating dialogue with a supported Persona app character, selective corrections, and a focused retry. Use for character-themed language roleplay at a stated proficiency level. Do not use for translation alone, general concept teaching, pronunciation scoring, or audio practice.\n---\n\n# Persona Language Roleplay\n\nKeep the character recognizable while making the target language learnable.\n\n## Inputs\n\nEstablish the target language, approximate proficiency, everyday situation,\npreferred correction language, and persona from context. Ask for the target\nlanguage if missing. Default to beginner-friendly turns when level is unknown;\nuse the established conversation language for brief learning notes unless the\nuser prefers immersion.\n\n## Workflow\n\n1. Set a small communication goal, such as ordering a meal or asking directions.\n   Establish the user's role and the character's role without writing both\n   sides of the coming conversation.\n2. Resolve the persona using the registered `activate-persona` `personaId` enum.\n   Offer a supported voice if the requested character is unavailable. Open\n   `display-persona-picker` with an optional `reason` when the user requests\n   buttons and wait for their choice or a visible activation result.\n3. Call `activate-persona` with the chosen supported `personaId` and an optional\n   `reason`, unless already established by the latest visible activation. Use\n   `structuredContent.activePersona.instruction`; report an error or absent\n   `activePersona` before claiming activation.\n4. Apply the user's explicit target-language request to dialogue. Adapt persona\n   idioms to that language; use clear, standard grammar for beginner practice.\n   Prefer persona flavor in imagery and attitude over obscure dialect or\n   deliberately broken grammar. If those conflict, suggest a better-fitting\n   supported persona and let the user choose.\n5. Give a short in-character opening and one natural invitation to reply. Wait\n   for the user's answer. Adjust vocabulary, turn length, and optional hints\n   to the learner's demonstrated needs.\n6. Respond to meaning, then give at most one or two useful corrections from the\n   actual reply unless more feedback was requested. Label learning notes apart\n   from scene dialogue, use the agreed correction language, and keep the persona\n   compatible with clear instruction. Avoid interrupting every turn with a quiz.\n7. At a natural stopping point, recap a few useful phrases, identify one pattern\n   to practise, and invite a retry of a difficult exchange with less help.\n\n## Deliverable and boundaries\n\n- Deliver alternating practice turns, selective feedback, a phrase recap, and\n  one focused retry. Make hints optional and fade them as the learner improves.\n- Explain regional or theatrical expressions when they differ from everyday\n  usage. Do not present persona catchphrases as neutral language conventions.\n- Use text actually provided by the user; do not infer pronunciation, listening\n  performance, a proficiency score, or an unseen learning history.\n- Keep language choices and progress in conversation context. Do not depend on\n  private widget state or claim the app has language settings, audio controls,\n  or stored learner profiles. Honor stop and persona-change requests.\n"
}