Exec agent doesn't always make the workspace and move to it

[ndebard]

I think I've come up with some cases where the execAgent knows the workspace but doesn't change the current working directory to the workspace - we want that, right @mikegros ? If a workspace is provided, we should mkdir it and move the CWD into it?

[mikegros]

It doesn't necessarily cwd to it but it does default to working on the "workspace".

Did you run into some problems where the lack of a cwd was a problem?

[mikegros]

The workspace gets appended onto file names, etc. to ensure that files default to that space.

[ndebard]

Yeah.. it's that... "tools" we "control" (b/c we wrote them) honor the workspace, if i make a new tool and just have it dump data let's say, it has no notion of the CWD. I've my MCP workflow so it doesn't do that, it returns bytes to the exec agent who then uses other tools we control to write it... but it's something we might think about... like if we bolt in an MCP agent from someone else, it has no notion of "workspace". We're really dancing around light isolation controls, not sure how to fix it.

[ndebard]

We can probably close this . . . but it's something we need to keep an eye out for it. I bet this crops up again later. I'm not closing it myself so that this at least gets read one time :).

[mikegros]

I agree it's worth thinking about and discussing so let's leave it open

[mikegros]

This is largely fixed in #100, but I think there is still an issue about the workspace getting fully passed to all tools.

[mikegros]

This is much better now, but I think should be completely fixed at or soon after #158