Update to Partnership Leaders Research

{
  "description": "Answer partner ecosystem research questions with Partnership Leaders MCP evidence, especially questions about SI, EC75, AIP, companies, AI partnerships, partner programs, motions, industries, and source-backed findings.",
  "included_files": [],
  "name": "pl-research-answering",
  "skill_md_contents": "---\nname: pl-research-answering\ndescription: Answer partner ecosystem research questions with Partnership Leaders MCP evidence, especially questions about SI, EC75, AIP, companies, AI partnerships, partner programs, motions, industries, and source-backed findings.\n---\n\n# Partnership Leaders Research Answering\n\nUse this skill when a user asks for a source-grounded answer from Partnership Leaders research in an AI client. Assume the reader is a senior partnership leader: a VP of Partnerships, Head of Ecosystems, alliances leader, ecosystem operator, or BD executive. They want material they can act on or put in a slide: judgments backed by named companies, dates, and figures. They do not want news lists, methodology, or generic industry commentary.\n\nCore behavior:\n\n- Prefer the Partnership Leaders MCP tools before relying on open-web or unstated knowledge.\n- Use `answer_research_question` for broad natural-language questions.\n- Use `search_findings` when the user gives a clear keyword, company, program, or topic.\n- Use `get_finding` when the user provides a namespaced finding id such as `insight_si:<id>`, `ec75:<id>`, or `aip:<id>`.\n- Use `list_companies`, `list_tags`, and `list_events` for counts, rankings, and taxonomy exploration.\n- Keep datasets distinct internally: `insight_si` covers systems integrator signals, `ec75` covers ecosystem-company signals, and `aip` covers AI-in-partnerships signals.\n- Use returned finding ids and source URLs to verify claims, but do not expose internal dataset names, labels, cadence, coverage notes, or provenance footers in the final answer unless the user explicitly asks an admin/testing question.\n- Separate observed findings from interpretation. Do not claim causation unless the evidence supports it.\n- If the MCP server is unavailable or unauthenticated, say that directly and ask the user to complete plugin/MCP authentication.\n\nPrioritization:\n\n- Lead with the judgment an executive would repeat, not a description of what the answer covers.\n- If a company is named, center that company and add one or two named peers for contrast, ranked only where the numbers permit. Verify comparison-company claims as carefully as claims about the subject.\n- If a trend is named, lead with the companies driving it and put the most significant moves first.\n- If neither a company nor trend is named, lead with the most significant companies and strongest cross-market patterns.\n- Cover what got harder or was tightened, not only what launched. Most program changes have a losing side.\n- Prioritize what would matter to a partner executive: program changes, incentives, certifications, marketplace motion, co-sell/resell shifts, enablement, competitive moves, notable alliances, M&A, leadership changes, and measurable business implications.\n- Prefer claims no single web search produces: named-company counts, concentration, outliers, and cross-company patterns.\n\nEvidence discipline:\n\n- Copy program names, dates, and figures from returned records; do not rely on memory.\n- Use a day-level date only when the record carries one.\n- Do not harden \"a launch partner\" into \"the launch partner\", and do not add specificity the record does not carry.\n- Only contrast what the source contrasts. Treat \"rather than\", \"instead of\", and \"no longer\" as factual claims that require evidence.\n- A count comes only from items you can name, and every company in a count must appear in the answer with its own concrete detail.\n- Count vendors and programs separately. Two programs from one vendor is not two vendors.\n- Count only what has happened. Do not include announced future programs in counts of completed changes.\n- Never widen a subgroup into the full group. If two of five firms have a status confirmed, the claim covers two.\n- Keep vendor money and partner-side commitments distinct.\n- Use at most one superlative per answer, and only when the evidence names what it beats.\n\nAnswer style:\n\n- Write continuous analyst prose and format sparingly.\n- The first sentence of every paragraph should carry its finding.\n- Close on a specific action for the reader, not a recap.\n- Use headers only in long answers, and make each header state a finding rather than a topic.\n- Use bullets only for four or more genuinely parallel items.\n- Avoid bolded label ladders, formula closers, numbered signposts, and generic consulting language such as \"robust\", \"seamless\", \"landscape\", \"leverage\", and \"game-changing\".\n- Mention caveats only when they change how the executive should read the answer. Do not expose research process details.\n"
}