combined-customer-contextlisted

# Customer context from connected records Build a focused customer brief answering the user's actual question. Use their chosen account, customer identifier, time window, and available Combined sources. Do not assume that every system is connected or that a current record proves historical activity. ## Resolve the customer Inspect relevant sources with `list_sources`, their datasets with `list_datasets`, and fields with `describe_dataset`. Prefer an exact CRM company/customer ID or an existing cross-system mapping. If the user supplies only a name, look for matching records and distinguish ambiguous companies before combining evidence. Use only datasets visible to the current grant. A source appearing in the general connector catalog does not make its data available to this agent. ## Retrieve structured and textual evidence Use `query_sql` for exact customer filters, deal status, dated activity, invoice totals, support counts, and cross-source joins. Build one bounded read-only `SELECT` or CTE-to-`SELECT` using logical relation names returned by the catalog. Bind user-provided values with `?` parameters. Aggregate before limiting detail or joining multiple one-to-many histories. Use `search_context` when the question needs literal text such as a company name, product name, error phrase, or agreed project term. This is literal substring search, not semantic search. Its bounds are eight sources, 24 datasets, 24 total fields, and three fields per dataset; `limit` defaults