A city clerk needs an agent that uses filing instructions from a knowledge store and looks up current fees from an API. How should those capabilities be exposed?

A city clerk agent that uses filing instructions from a knowledge store and looks up current fees from an API should expose both capabilities as tools: a knowledge-store tool for instructions and an API tool for fees. Foundry agents compose knowledge retrieval for stable procedural text with function-calling for volatile authoritative values. Filing steps belong in the knowledge store; today's fee amounts belong in the fee API so the clerk desk never quotes stale numbers baked into prompts. Using a cached API tool for fees keeps the knowledge-store wording for instructions but risks quoting stale fee amounts from cache instead of calling the live fee API for current fees. Integrating an API tool for cached fees makes the same staleness mistake on the fee side, surfacing out-of-date figures held in cache rather than the live amounts the fee API returns, which is exactly the drift a live fee tool prevents. Using an indexed API tool for fees treats volatile fee values as pre-indexed searchable content rather than a live lookup, so the agent can return an indexed snapshot rather than today's authoritative amount from the fee API. On the exam, combine knowledge-store and API tools for clerk workflows; distractors keep the filing-instructions wording but weaken the fee lookup as a cached API tool, an API tool for cached fees, or an indexed API tool rather than the live fee API. Caveat: tool routing still needs clear descriptions so the model calls the fee API for amounts and the knowledge store for procedures—ambiguous tool text causes mixed-up sources. Operational check: ask a filing question that requires both a procedural step and a fee, verify the knowledge tool and fee API are both invoked, and confirm the quoted fee matches the live API—not the system prompt or the knowledge store.