How do you extend HF Agents with new tools?

• A. Create a separate execution environment and forget to register.
• B. Implement a new Tool wrapper, document it, register it in the registry, and update prompts and plans.
• C. Directly call external APIs without wrapping.
• D. Replace all existing tools with the new one.

Answer: (B)

To extend HF Agents with new tools, you wrap the tool in a standardized Tool wrapper, document it, register it in the tool registry, and update the prompts and planning logic so the agent knows how and when to use it. The Tool wrapper defines how to call the external capability, how inputs are mapped, how outputs are parsed, and how errors are handled, ensuring consistent interaction with the agent’s existing tool-using flow. Documenting the tool provides clear usage details, inputs, outputs, and any constraints, which helps maintainability and onboarding. Registering the tool makes it discoverable at runtime, so the agent can access it just like other tools instead of hardcoding or duplicating functionality. Updating prompts and plans teaches the agent about the new capability, including its input schema and examples of when to invoke it, so the planning component can incorporate it into its reasoning. Separately executing APIs without wrapping, or replacing all tools, would break standardization, discovery, and safety, and would make the agent harder to manage and extend.