Repository navigation
Support skills #241
Description
To upvote this issue, give it a thumbs up. See this list for the most upvoted issues.
We have a recent protocol for skills used by claude, cursor and others, we should discuss how to support in eca complementary to existing features.
I'm interested in use cases that current ECA's features do not support so we can understand how it fits, please share if you have one
Activity
Ramblurr commentedon Dec 22, 2025
For completeness: there's some discussion in Slack https://clojurians.slack.com/archives/C068E9L5M2Q/p1766086833969739
zikajk commentedon Dec 30, 2025
One thing -> Let's say that config has a defined path to the skills folder (where ECA reads and creates a system prompt from them during initialization).
It is not uncommon for someone to prepare skills using LLM. At that moment, the system prompt could be modified under their hands (and invalidate the cache), so I would only load the skills once, at the start of the chat and make sure that hot reload won't be applied to the stuff that could modify system message. WDYT?
zikajk commentedon Dec 30, 2025
Btw. OpenCode has official support for Skills now:
anomalyco/opencode#5930
anomalyco/opencode#6000
I think it is a good idea to define them inside behavior (or globally). But I am not sure if we want extra layer of permissions for them.
Althought It would be easier than having to configure them via
argsMatchers.allowed-toolsshould really work.ericdallo commentedon Dec 30, 2025
Thanks for the info @zikajk! I still need to understand better the feature, probably asking LLM how opencode works on that hehe, but I can see there is a native skill tool for that.
I think we can start supporting the basic, and config still via argsMatchers and understand if we need finer tuning/config for skills WDYT?
Skills support following https://agentskills.io.