GitHub user zyratlo added a comment to the discussion: Proposal: Making the workspace agent useful to a new user by seeing run errors, reading the docs, knowing the datasets
Thanks for writing this up! This lines up with what I saw with my collaborator, and I think all three are worth doing. I have one comment on 1. Like I mentioned in #9246, my collaborator ran into runs that finished with no errors but still gave wrong results, like an empty dataset giving an empty result. If only fatal errors and error-level console messages get attached, the agent still can't answer "why is my result empty?" Could the run summary also include something like each operator's output row count from the user's latest run? Then the agent could see that, for example, a file scan produced 0 rows and point at the dataset. The agent already reads row counts for the runs it starts itself, so it's the same information coming from the user's run instead. For 2, I agree on bundling. I'd expect the list of questions with no docs page to be the most useful part for #8773. A lot of my collaborator's questions weren't about building workflows at all (finding a shared workflow, selecting a file in a dataset, export failing because a dataset wasn't downloadable), and I checked `docs/` and none of those are covered right now. Most of the pages are operator references or about installing and contributing, and only a few are written for end users. So for a new user, the docs tool will only help as much as those pages get written. I can share the questions they asked if that helps as a test set. No objection on 3 right now. GitHub link: https://github.com/apache/texera/discussions/9377#discussioncomment-19500045 ---- This is an automatically sent email for [email protected]. To unsubscribe, please send an email to: [email protected]
