So it looks now like the automations are getting c...
# proj-automation
f
So it looks now like the automations are getting confused between webhook based invocation and scheduled invocation - for me it is now setting up polling automations but not scheduling them and expecting them to be invoked by webhook
e
for these polling-based automations, is OpenHands generating the python script, or is it using something pre-built? Given the creation of automations is LLM-driven, does choosing an underpowered model mean you could get a 'broken' polling script that doesn't actually meet the user requirements? Basically, I wonder if these pre-built automations have the potential to be very brittle.
f
The issue I am facing now is actually with skills - the agent is not finding the automation skill and trying to build from scratch, and coming up with janky results
So far I've seen it think that the webhook approach should be used - create an automation which does not have a schedule It also tries to create the automation from scratch
nod 1
If I does use the skill, it gets a script which is very specific
I see there have been some changes to the skill since I authored it, which may be responsible for this
e
I will say, I think nailing this is important - a user journey goal for Agent Canvas is to get users to activate an Automation quickly and reliably, so we want to make sure this is a top notch experience
f
It is my primary focus today
❤️ 1
thanks on a stick 1
• WORKING: Updating agent canvas to include breaking changes in the extensions repository • TODO: Fixing issue where skills are not picked up by the agent • TODO: Refining the process so the agent does not get confused between webhooks and polling