GitHub Copilot can load plugin skills, but cannot load any resource internal to the skill

[gsemet]

🏷️ Discussion Type

Bug

💬 Feature/Topic Area

VS Code

Body

I have several skills that are provided by a Plugin on a custom marketplace. I can see the project successfully cloned to ~/.vscode/agent-plugins.

In VS code, I can trigger it using slash-command, but I saw that the model complained it could not trigger this skill based on the intent ("I do not see this skill"). That is so problematic, because in the Agent Customization panel, I can see the skill available.

That is not the worst problem. The problem is, when called, the file SKILL.md is loaded into the context, and then the model CANNOT load any relative file by progressive disclosure (like a relative markdown files, a template, a image,...)

That makes using skills by marketplace very limited to simple skills that does not rely on progressive disclosure. This is not how all the other coding agent works.

Thanks

[github-actions[bot]]

💬 Your Product Feedback Has Been Submitted 🎉

Thank you for taking the time to share your insights with us! Your feedback is invaluable as we build a better GitHub experience for all our users.

Here's what you can expect moving forward ⏩

• Your input will be carefully reviewed and cataloged by members of our product teams. 
  • Due to the high volume of submissions, we may not always be able to provide individual responses.
  • Rest assured, your feedback will help chart our course for product improvements.
• Other users may engage with your post, sharing their own perspectives or experiences. 
• GitHub staff may reach out for further clarification or insight. 
  • We may 'Answer' your discussion if there is a current solution, workaround, or roadmap/changelog post related to the feedback.

Where to look to see what's shipping 👀

• Read the Changelog for real-time updates on the latest GitHub features, enhancements, and calls for feedback.
• Explore our Product Roadmap, which details upcoming major releases and initiatives.

What you can do in the meantime 💻

• Upvote and comment on other user feedback Discussions that resonate with you.
• Add more information at any point! Useful details include: use cases, relevant labels, desired outcomes, and any accompanying screenshots.

As a member of the GitHub community, your participation is essential. While we can't promise that every suggestion will be implemented, we want to emphasize that your feedback is instrumental in guiding our decisions and priorities.

Thank you once again for your contribution to making GitHub even better! We're grateful for your ongoing support and collaboration in shaping the future of our platform. ⭐

[Amorfati123]

Thanks for writing this up. I think this is an important limitation, especially for marketplace-provided skills that are meant to be more than a single static prompt.

The main issue seems to be that the skill is visible and callable at the UI level, but it is not fully available as a usable workspace-like resource to the agent. Loading only SKILL.md is not enough if the skill depends on supporting files such as examples, templates, schema files, reference markdown files, or images.

I would expect a skill package to support something like progressive disclosure, where SKILL.md gives the top-level instructions and the model can then read related files from the same skill directory when needed. Without that, skill authors are forced to put everything into one large SKILL.md, which is harder to maintain and wastes context.

A few things that would make this much more reliable:

1. The agent should know the skill’s root directory when the skill is invoked.
2. Relative links or file references from SKILL.md should resolve against that skill root.
3. The model/tooling should be able to read additional files from the skill package on demand.
4. If relative files are blocked for security or sandboxing reasons, the UI should make that clear and document the supported pattern.
5. If the skill appears in the Agent Customization panel, the intent router should also be able to discover and select it consistently, not only through explicit slash commands.

The current behavior makes custom marketplace skills useful only for very simple cases. For larger skills, especially ones with templates, reusable examples, or structured documentation, this prevents the skill from working the way most authors would expect.

[gsemet]

Thanks for confirming i am not the only one to suffer from this issue. There is also the LFS support that is not good (i store svg and other reference files as LFS files). and during clone, despite the presence of the executable git-lfs in PATH (obviously), it still clones without the LFS, leaving a state with bad clone and invalid path...

[Amorfati123]

Yes, that makes the problem even worse. If the plugin clone leaves LFS pointer files instead of the real assets, then even a future fix for relative file access would still fail because the referenced files are not actually present.

So there are really two separate issues here:

1. the agent cannot reliably resolve/read relative files from the skill package
2. the marketplace/plugin clone does not fully hydrate the repo when LFS files are used

For skill packages, I think GitHub should either run the equivalent of git lfs pull during install/clone, or clearly document that LFS assets are not supported. Otherwise the skill can appear installed correctly while being broken internally..