draft for workflow design #3537
Conversation
| llm: NotGivenOr[llm.LLM | llm.RealtimeModel | None] = NOT_GIVEN, | ||
| tts: NotGivenOr[tts.TTS | None] = NOT_GIVEN, | ||
| allow_interruptions: NotGivenOr[bool] = NOT_GIVEN, | ||
| extra_instructions: NotGivenOr[str] = NOT_GIVEN, |
There was a problem hiding this comment.
for these built-in workflows, I think we should allow users to add extra instructions or even modify instructions completely. for a real productions, users may want to add a lot of other prompts like context, tones, fallback handling, examples…
| self._email_update_speech_handle: SpeechHandle | None = None | ||
|
|
||
| @property | ||
| def base_instructions(self) -> str: |
There was a problem hiding this comment.
besides adding a extra_instructions option, how about expose base_instructions as a property, so users can modify it by creating a sub class if they want.
| denied_reason: str | None | ||
|
|
||
|
|
||
| class CollectConsentTask(AgentTask[CollectConsentResult]): |
There was a problem hiding this comment.
CollectConsent might be another common task (e.g. #3089), it will require more customization on instructions to add the context
| ) | ||
|
|
||
| @property | ||
| def base_instructions(self) -> str: |
There was a problem hiding this comment.
Maybe we can override instructions?
There was a problem hiding this comment.
the original instructions property is for users to read the instructions, we won't set the instructions from it?
No description provided.