Thoughts on SDK Abstraction for Extensions

[ognis1205]

Hi all,

After reviewing the A2A Extensions guide, I had a few questions and thoughts regarding how SDKs might evolve to support them in practice:

1. SDK Abstraction for Method Extensions

• Currently, most SDKs provide the core A2A methods in a type-safe way (e.g., client methods).
• However, new RPCs introduced by Method Extensions cannot be exposed under the same abstraction.
• Should SDKs consider a mechanism like client.add_extension(MyExtension) where an A2AExtension interface can inject new methods dynamically?

2. Discovery & Bootstrapping

• In more autonomous client environments, discovery may yield agents with extensions not known ahead of time.
• This is especially relevant for Method Extensions: for extensions marked as required: true, does this introduce a bootstrapping-time issue where the client must already have the implementation available before it can interact with the agent?

3. Naming Collisions

• If multiple extensions define overlapping method names or data object names, do we need a namespacing mechanism (e.g., extensionName.methodName)?

4. Spec Publication vs. Implementation

• The current draft does not specify how extension specifications should be linked to their reference implementations.
• Should there be a convention (e.g., extension URL -> SDK package) to avoid fragmentation?

Curious to hear if others have thought about these points or if there are existing design directions around them.
Thanks!