agent-shell can't run for the first time when not able to connect to to githubusercontent.com and avatar.github.com

[cowboyd]

I'm running agent-shell in a "behind-the-firewall" application and it's failing to populate the cache of the lobe-icons because those urls are not on the whitelist. I have to work manually download them, carry them through the firewall and then copy them into the cache directory so that agent-shell never makes the attempt to retrieve them.

It would be awesome if there was an "asset bundle" that you could point to and it would copy them into the cache from there instead of dynamically fetching them from the internet.

Ideally, there would be a single tarball we could use, maybe built into the releases from GitHub that could serve as a single download. That would make manually copying a lot easier, and it would also mean that if we do go to our IT departments to ask for an exception, it is easier to get one for a single url.

Checklist

• I agree to communicate with the author myself (not AI-generated).
• I've read the README's Filing issues section.
• I'm running the latest versions (fill in below).
  • agent-shell version:
  • acp.el version:
  • ACP package (e.g. claude-agent-acp) version:
  • Agent CLI (e.g. claude, gemini) version:

[xenodium]

Thanks for filing.

Can you share the error you ran into? Got a stack trace? Something else?

While asset bundling sounds nice in principle, it it will generate a bit more maintenance work, which unless I get many more requests for this feature, may not be worth the squeeze.

I'm wondering if you'd be ok with some compromise here? If there's an error downloading assets, we could use fallback images, but remain functional. We already do a similar thing for models that don't supply a download URL (we generate an image per model).

[cowboyd]

@xenodium It was awhile back, so not sure if I have a stack trace, but if memory serves, it was a DNS error because the dev machine has limited access.