What do you want to change?
Add an optional streamingBehavior?: "steer" | "followUp" field to extension input events when the prompt was submitted through session.prompt(..., { streamingBehavior }).
Why?
Input handlers can currently see raw text/images/source before queueing, but they cannot tell whether a streaming prompt came from Enter (steer) or Alt-Enter (followUp). Extensions that intercept input and re-submit paired messages need to preserve the user's queue intent without guessing.
How? (optional)
Thread the existing PromptOptions.streamingBehavior through AgentSession.prompt(...) into ExtensionRunner.emitInput(...) and include it on InputEvent.
The field would be optional, undefined for idle prompts, and would not affect direct session.steer() / session.followUp() because those bypass input today. Add tests in packages/coding-agent/test/extensions-input-event.test.ts and update the input-event docs.
What do you want to change?
Add an optional
streamingBehavior?: "steer" | "followUp"field to extensioninputevents when the prompt was submitted throughsession.prompt(..., { streamingBehavior }).Why?
Input handlers can currently see raw text/images/source before queueing, but they cannot tell whether a streaming prompt came from Enter (
steer) or Alt-Enter (followUp). Extensions that intercept input and re-submit paired messages need to preserve the user's queue intent without guessing.How? (optional)
Thread the existing
PromptOptions.streamingBehaviorthroughAgentSession.prompt(...)intoExtensionRunner.emitInput(...)and include it onInputEvent.The field would be optional, undefined for idle prompts, and would not affect direct
session.steer()/session.followUp()because those bypassinputtoday. Add tests inpackages/coding-agent/test/extensions-input-event.test.tsand update the input-event docs.