Most tools have clearly distinct resource+action targets, and the descriptions proactively disambiguate (e.g. calculate_quote 'without saving', suggest_group 'not an availability check'). The weakest boundary is the read cluster get_events / get_calendar / get_objects, which all describe public schedule/resource information and could be confused by an agent, as could calculate_quote vs suggest_group for advisory tasks.

Every tool follows a strict verb_noun snake_case pattern (get_events, calculate_quote, save_estimate, submit_application). The estimate lifecycle uses a coherent get/save/edit triplet, making the convention highly predictable.

12 tools sit comfortably in the well-scoped range, and each maps to a distinct stage of the workflow: discovery, quoting, estimate drafting, and application submission. No tool looks like filler or a duplicate alias.

The surface covers discovery, quoting, estimate CRUD (save/get/edit), and application submit/get, but there are notable gaps: no tool actually checks availability despite repeated disclaimers that other tools do not, and there is no cancel/withdraw or update path for a submitted application, nor a list/delete for estimates. Agents hitting these dead ends will have to fall back on contacting the organizer.