library/processes/shared/db-agent-explore overview

Generic "point an analyst agent at a local SQLite DB and ask a research question" process. Generalized from the query-composition + narration phases of `specializations/domains/business/travel/travel-plan-compose.js`. The analyst reads SCHEMA.md (no guessing), composes SQL via a Python 3 stdlib `sqlite3` script, executes it (read-only, `mode=ro` URI), iterates until the findings answer the question, and writes a human-readable report with verbatim SQL as audit evidence. Hard constraints: - All tasks are `kind:'agent'`. No shell tasks. No MCP. - DB is opened read-only so the analyst cannot mutate the dataset. - Every finding carries the verbatim SQL that produced it.