zozo123 commented on issue #74194: URL: https://github.com/apache/airflow/issues/74194#issuecomment-6073379181
Thanks @garvit-arora, that slice is yours. Hook + managed-agent toolset adapter first, resumable standalone tasks later, exactly as scoped. A few things that will make review quick: - Build the adapter on the public `BaseManagedAgentToolset` contract from #73532 (`common/ai/toolsets/managed_agent.py`) and don't override private common.ai internals. Keep the adapter thin and separate from the hook, so the hook stands on its own and common.ai stays an optional extra. - Starting a conversation or posting a message is not idempotent. On an ambiguous failure (timeout, 5xx after send) surface an error rather than retrying, and only retry reads (get message, fetch result). - Bound result size and keep tokens out of anything serialized. Return space/conversation/message IDs plus a truncated, structured result. - Map 401/403/404/429 to clear errors. Note that Databricks sometimes answers a missing resource with 403. - Docs: say when the managed Genie MCP server via Unity Gateway is enough, and what the native API adds. - Please include one real-workspace run (PAT is fine) recorded separately from the mocked tests. My review on #74198 covers several of the same pitfalls if you want a checklist. Happy to review once it's up. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
