@Vince 500 error submitting an agent observation example - i think there's too much data associated with the getActivityRhythm tool (at least that's my hypothesis because this is the first time I called it with 30d)

you're in staging yeah?

yeah

bumped to 5mb for the agent examples, and pushed to staging, if you wait a minute or two it should work

is it surprising that the tool call would be using that much?

that makes me nervous in terms of server memory

the full saved example includes all the tool calls for the example (and the original response), so if you have a couple tool calls with long timelines it can get big (particularly the user timeline stuff with all the <1m sessions)

I think it's ok but we should definitely keep memory allocation in mind when it comes to these large tool calls. That would be great. Also just a reminder to use TOON when sending these to the LLM

they are in fact being TOON'd for the agent, but when they're run through the admin console routes we need to be able to parse them for formatting and rendering nicely so they stay in json

this is for staging dashboard.render.com/web/srv-d6ng445m5p6s73csg0d0/metrics?u=total&amp;r=status_code

@Vince the agent should do a policy audit. When a policy gets deleted that should always be mentioned and there should be like a several day “reminder” of some sorts for observation agent

got it, there's currently just the last 24 hours in its prompt but I will bump that up and make sure it's highlighted properly in the prompt