Skills Reference
Skill Management
What it does
Creates and deletes custom managed skills. The meta-skill that lets you extend your assistant with new capabilities.
Setup required
None. Works immediately.
Permissions
- Skill creation/deletion is classified as high-risk and always requires approval
Common prompts
| You say... | What happens |
|---|---|
| “Build me a skill that checks Hacker News for trending posts” | Scaffolds a new custom skill |
| “Create a skill for managing my project's deployment pipeline” | Creates a domain-specific skill |
| “Delete the Hacker News skill” | Removes a custom skill |
| “Show me my custom skills” | Lists managed skills |
Configuration
- Creates a skill folder: SKILL.md (instructions) plus optional companion files, such as references/*.md for failure modes and cached values, and scripts/* for reusable code the skill runs (invoked with python3 or bun)
- Companion scripts are plain files the assistant runs from the terminal; skills built this way cannot register executable skill tools (TOOLS.json ships only with built-in and community-installed skills)
- Skills are saved to ~/.vellum/workspace/skills/
Tips & gotchas
- Extend your assistant. Custom skills are how you add capabilities beyond the built-in set.
- Fully automated.The assistant writes the code, tests it, and packages it — you just describe what you want.
- High-risk classification. Skill source file modifications always require approval to prevent privilege escalation.
- Review before approving. Check the generated code before granting approval. You can also edit skill files directly if you prefer.
- Scripts come from proven runs. When the assistant stores a scripts/ file in a skill, it saves the exact code it already ran successfully, so reusing the skill later runs the script instead of rewriting the code from scratch.