What problem does it solve? Row-at-a-time filters and BM25/vector retrieval cannot express relational or aggregate people-search questions like career ordering, tenure overlaps, or network-wide statistics. This Skill answers those questions by running read-only SQL directly against the local search DuckDB. ## Core Features & Use Cases - Per-person aggregates: Answer questions like "2+ stints at seed-stage companies" or "average tenure under 2 years" using GROUP BY and HAVING over position data. - Person-to-person joins: Find people who overlapped with a target at the same company, schoolmates, or set algebra across sub-populations via self-joins on date ranges. - Network analytics: Produce aggregate tables about the network itself, such as which companies you know the most people at or how many contacts changed jobs recently. - Use Case: A recruiter asks "who can intro me to John Doe?" The Skill resolves John Doe's companies and date ranges, then runs an overlap self-join to find shared-tenure colleagues, clearly flagging the result as overlap-based inference. ## Quick Start Ask the agent to use search-sql to find everyone in my network who overlapped with a specific person at the same company. ## Schema & Method The Skill ships a schema cheat sheet covering local_people_positions, local_person_profiles, local_summaries, local_people_education, local_education, and local_companies. It enforces value-space probes before filtering on categorical columns, dedupes results to person grain, caps output at 100 people, and requires the final SQL plus coverage caveats in every answer.
