Feature Request Currently • Slack conversation sends to the user a response to an application-level automatic "summarize" prompt (sent as a "user" message) • it doesn't send to the user the actual agent's response to the question • if I asked a question, I find that what I actually want is the response, not some summary of what the agent did (albeit some will be embedded anyway in its response) • at least with LLMs that follow instructions faithfully and don't forget them / overwrite them, like GPT-5.x, I found out that it works well to prompt the LLM from the initial Slack message like "when I ask you to summarize, don't summarize, just repeat to me verbatim what you just told me" or some variation Maybe it's better: • if the user asked a question, send to Slack the actual response • do not insert the summarize prompt, idk, it feels like noise in that case?

I think it also make sense to just send the last agent finish message to the user and potentially remove the summary prompt entirely