A document analysis subagent has a generic fetch_url tool. It sometimes loads search result pages and blog posts instead of the source papers it was given. What is the best fix?

Answer: B — Replace fetch_url with a load_document tool that validates the URL points to a document

A constrained tool that only accepts valid document URLs makes the misuse impossible while still covering the agent's job (b). A description warning (a) reduces but doesn't prevent misuse. Post-filtering (c) still spends the call and the context on the wrong page. Coordinator approval (d) adds a round trip to every fetch to work around a tool that is too general.