Post

Interesting! Is the example specific to a harness or can it be used for any tool definition?

this is super cool! It definitely reminds me of deferred tools, but for a model without specific training for it. It also sounds like it can be implemented inference side, to prevent needing a round trip as well. I really like the ack as well

I think that using this to keep the system prompt small could lead to the model ignoring the guidance in tool call output due to the instruction hierarchy and anti-jailbreak. Deferred tools schemas are appended as system/developer messages iirc

Guidance injection at inference would be awesome. However, the guidance my app injects is user-editable at runtime. I think we'd lose that if this was inference side?

it should definitely be possible! deferred tools work like this inference side too to prevent round trips. As long as you know the tool->guidance mapping at api request time (chat completions, responses api, etc), you can just send both to inference and inference can manage the guidance flow