This message was deleted.
# proj-agent-canvas
d
This message was deleted.
w
0 triggers means "I always want the agent to know this"
c
Right
w
so I'd amend your suggestion of
• don't require any type specified, use
triggers
to recognize type
knowledge
and say
• don't require any type specified, use
triggers
to recognize "is summarizable"
• and get rid of type completely
but loosely held opinion
c
I should have said it's not summarized by default, not quite that it cannot be. Currently, it's an effect of setting the events we always keep to 3: which will always include repo
w
ohh interesting
yeah that makes sense
I think that'd "just work" because any zero-trigger microagent will end up in that first event (do they get combined together in one event?)
for some more context, I want to add a "user" microagent that behaves exactly like "repo"
c
Yes, it's the 3rd event strictly speaking (from the condenser perspective): user message, recall action, recall obs. The recall obs will concatenate all
repo
microagents contents
šŸ‘ 1
w
always on, but for me personally rather than anyone working with the repo
c
That seems to imply: • load microagents from a user-defined path • and have, hah, a... repo name property , for the repos it applies to It wouldn't be in git, right?
Or is it acceptable to be loaded from a user-defined path, and applied to all repos?
w
my thought is that I'd have a
rbren/.openhands
repo with all my OpenHands customization, including my "user" microagent and any trigger microagents I might want to define
šŸ¤” 1
šŸ‘€ 1
(stepping back, I think prompts-as-code is a really good practice we should promote)