GitHub user bobbai00 edited a comment on the discussion: Proposal: Making the 
workspace agent useful to a new user by seeing run errors, reading the docs, 
knowing the datasets

Thanks @eugenegujing for the clear write-up and the careful PRs.

I just wrote a task regarding agent refactoring (#9453). The goals are aligned 
here: the agent should see the user's executions and answer from the docs. What 
overlaps is how each PR gets that information into the agent, which is exactly 
what the refactor in #9453 changes. Here is how I see them fitting in:

1. **#9423, my suggestion is to hold this one for now**: the agent should join 
the workflow's shared editing and execution session, so it gets the same 
execution events as the user's canvas, without a report from the browser. Its 
rules for explaining errors and reading row counts fit nicely as a skill.
2. **#9424, my suggestion is to hold this one for now**: the docs search & 
related tools should become skills and MCPs. The agent sees the list of user 
pages and reads the one it needs, so we don't need a separate search index. The 
25 questions are a great start for the evaluation set we should add to agents, 
together with the questions @zyratlo offered.
3. **#9426, I approved**: this one is a good feature.



GitHub link: 
https://github.com/apache/texera/discussions/9377#discussioncomment-19640296

----
This is an automatically sent email for [email protected].
To unsubscribe, please send an email to: [email protected]

Reply via email to