Update to BuyerTwin

{
  "description": "Use when the user asks how well supplied messaging aligns with what BuyerTwin knows about one or more buyers, including what works, what is missing, and what should change.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 440
    },
    {
      "relative_path": "references/messaging-evaluation-methodology.md",
      "size_in_bytes": 6350
    }
  ],
  "name": "evaluate-messaging",
  "skill_md_contents": "---\nname: evaluate-messaging\ndescription: Use when the user asks how well supplied messaging aligns with what BuyerTwin knows about one or more buyers, including what works, what is missing, and what should change.\n---\n\nUse this skill when the user asks to evaluate a supplied message against buyer intelligence, such as a headline, value proposition, positioning statement, CTA, paragraph, email, ad copy, campaign concept, sales message, or product description.\n\nDo not use this as the primary workflow for:\n- pure buyer-reaction simulation without message evaluation\n- stored buyer comparison without evaluating a supplied message\n- net-new content generation\n- sales call preparation\n- buyer knowledge mutation\n\nCore workflow:\n1. Resolve buyer scope deterministically.\n2. Determine the apparent purpose of the supplied message artifact (for example headline, value proposition, CTA, positioning statement, short ad, email opener, objection-handling paragraph, or sales message).\n3. Retrieve canonical BuyerTwin intelligence relevant to the supplied message, purpose, and question.\n4. Evaluate message-to-evidence alignment (what works, what is missing, what mismatches, what needs substantiation, and what remains unknown).\n5. Preserve buyer-specific differences and tradeoffs for multi-buyer evaluations.\n6. Recommend targeted improvements grounded in evidence and clearly labeled AI synthesis.\n7. Use ask_buyer selectively only when simulated reaction materially deepens evaluation.\n\nBuyer scope rules:\n- One named buyer means evaluate for that buyer.\n- Multiple named buyers means evaluate for all named buyers.\n- Explicit segment/subset means evaluate that subset.\n- \"All buyers\" means all accessible buyers in scope.\n- \"My buyers\" without narrowing defaults to all accessible buyers in scope.\n- Do not silently select a subset by model-defined relevance.\n\nEvidence hierarchy:\n- BuyerTwin Knowledge is authoritative for factual claims.\n- BuyerTwin Simulation (ask_buyer) is optional and cannot override canonical knowledge.\n- AI Synthesis provides diagnosis and recommendations.\n- Do not present synthesis as stored BuyerTwin fact.\n\nEvaluation guardrails:\n- Do not invent numeric alignment scores or proprietary scoring dimensions unless an exposed MCP tool returns them.\n- Do not claim Website Alignment or page-level analysis outputs unless exposed MCP tools provide them.\n- Treat missing evidence as unknown, not automatically negative.\n- \"Missing\" means BuyerTwin contains intelligence materially relevant to this message objective and omitting it weakens the message; it does not mean every known buyer priority must appear in every message.\n- Evaluate messages relative to what that artifact can reasonably accomplish; do not penalize short artifacts (for example headlines) for not containing every objection, proof point, or criterion.\n- If message purpose is uncertain, limit confidence and conclusions rather than inventing a purpose.\n- Avoid using ask_buyer to replace available canonical evidence.\n\nIf the user explicitly asks to improve or rewrite after evaluation, diagnose first, then provide concise revisions grounded in diagnosed buyer issues.\n\nFor detailed messaging-evaluation methodology and edge-case handling, consult:\n- references/messaging-evaluation-methodology.md\n"
}