I love Anthropic Claude Code (I have the Max subscription), but I use it so much that I regularly exhaust my weekly quota before the end of the week. When that happens, I usually switch to other free-tier CLIs, for example, Google Antigravity, GitHub Copilot, and, just last week, Mistral AI's Vibe. I really wanted to try Vibe to evaluate the feasibility of moving away from Claude Code in favor of a more sovereign solution (i.e., an European one).
Vibe only offers a single model by default, Mistral Medium 3.5. It hallucinated results, and they were so bad that I noticed it immediately. When I researched to try to understand what happened, I realized that it hadn't triggered any of the skills that I had previously crafted with much love. The reason is that Claude Code stores skills in custom locations, in opposition to most other assistants.
I went through each assistant's documentation, so you don't get bitten the same way and here is the result:
| Assistant | Personal | Project |
|---|---|---|
| Claude Code |
|
|
| Antigravity |
|
|
| Vibe |
|
|
| Copilot CLI |
|
|
The specification is pretty clear about it:
| Scope | Path | Purpose |
|---|---|---|
| Project |
| Your client’s native location |
| Project |
| Cross-client interoperability |
| User |
| Your client’s native location |
| User |
| Cross-client interoperability |
My conclusion is pretty straightforward. Put skills in ~/.agents/skills and ./.agents/skills, then symlink ~/.claude/skills and ./.claude/skills to them. If the .claude/skills folders already exist, move their content to the .agents/skillsones first.
ln -s ~/.agents/skills ~/.claude/skills ln -s ../.agents/skills .claude/skills # 1 - Run it from the project root
I definitely love Claude Code, but it isn't a good citizen here. It doesn’t abide by the specification. That's one more reason for me to try Mistral AI’s Vibe more seriously and try to definitely migrate from Claude Code to Vibe.
To go further:
