chat.agentFilesLocations and chat.agentSkillsLocations should support glob/wildcard patterns

[Jonas-K-S]

Summary

The chat.agentFilesLocations and chat.agentSkillsLocations settings currently only accept explicit relative paths or paths starting with ~/. Glob patterns using wildcards (e.g., ** or *) are not supported, making it difficult to work with multi-repo or workspace setups where agent/skill files are located at varying directory depths.

Problem

In a multi-repo workspace, agent and skill files may be nested inside sub-repositories or sub-folders at different depths. For example:

Currently, each path must be explicitly listed:

This is tedious and requires updating the settings every time a new sub-repo is added.

Expected Behavior

Glob/wildcard patterns should be supported, allowing a single entry to match agent/skill folders at any depth:

{
  "chat.agentFilesLocations": {
    "**/.agents/agents": true
  },
  "chat.agentSkillsLocations": {
    "**/.agents/skills": true
  }
}

This addition would be very helpful when working with Meta Repositories and Mono Repos.

VS Code Version: 1.113.0

[vs-code-engineering]

Thanks for creating this issue! It looks like you may be using an old version of VS Code; the latest stable release is 1.114.0. Please try upgrading to the latest version and checking whether this issue remains.

Happy Coding!

[Jonas-K-S]

This still applies to 1.114.0

[aeschli]

For performance reasons, we don't want to support global patterns.

Wouldn't you use a multi-room workspace for multi-repo repos?

The tedious task to update the settings every time a new sub-repo is added could be done by a script.
That would save you CPU cycles as for every LLM request we would have search every folder in your repo (to get the list of possible subagents)

[willyboy]

For performance reasons, we don't want to support global patterns.

Wouldn't you use a multi-room workspace for multi-repo repos?

The tedious task to update the settings every time a new sub-repo is added could be done by a script. That would save you CPU cycles as for every LLM request we would have search every folder in your repo (to get the list of possible subagents)

The March 2025 release notes explicitly mention support for glob patterns: https://code.visualstudio.com/updates/v1_99#_reusable-prompt-files

We built functionality around this feature that's documented but was apparently never implemented?

While I get the issue cites an extreme example, matching say, all scopes in node_modules for your company so that files are discoverable is a totally valid use case.

[aeschli]

The release notes were only for prompt files, where glob patterns are still supported but are not recommended for performance reasons.
I agree that finding files in non-modules is a good use case. If this becomes a standard/common pattern, we would like to support it (but without the need to add a general glob pattern)

[willyboy]

The release notes were only for prompt files, where glob patterns are still supported but are not recommended for performance reasons. I agree that finding files in non-modules is a good use case. If this becomes a standard/common pattern, we would like to support it (but without the need to add a general glob pattern)

We don't see this working for prompt files.

e.g.

"chat.promptFilesLocations": { 
    "a/b/c": true
}

finds a prompt file at a/b/c/d.prompt.md.

But

"chat.promptFilesLocations": { 
    "a/**/c": true
}

will not find that same file.

Is the official stance that none of this is supposed to work? We'll have to revert our solution if that is the case and I don't want to take that action if this is just an "Oops, we forgot to merge that feature"

[aeschli]

For it to work you have to use a/*/c/*.

But as said, it's marked as deprecated and support will be eventually be removed.