See if we can get the GitHub agent or an agent that uses GitHub for certain kinds of code reasoning to take the entire code into its context window using a utility similar to what the humans use with appropriate filters that are set dynamically. This would be an alternative to using ASTs

[jkbrooks]

Not sure we should actually do this, just considering it as an alternative. Probably we should prioritize ASTs first. I imagine we will actually get some smooth stuff, but we are definitely benefited by trying it out just to see the pros and cons. If we got downtime or someone else like external can try it out just so we know what would happen.

[jkbrooks]

This would require dynamically generating the filters for a utility like the ones here #75. Maybe even standardizing them and sharing them?

It would also involve dealing with the fact that Gemini does not have structured outputs the way OpenAI does. Right now, we enforce structured outputs on JSON that we send back and forth in the Eliza ecosystem. Without that, there is a chance you can get arbitrary failures. Gemini does not support this at all, which is why we haven't actually tried to use Gemini. Essentially, it's a non-starter to OpenAI. You have an Eliza agent use Gemini without addressing this issue one way or another.

[github-actions]

This issue has been automatically marked as stale due to 30 days of inactivity.
If no further activity occurs within 7 days, it will be closed automatically. Please take action if this issue is still relevant.