Attach skills
Skills are markdown files that load alongside an agent's prompt. Use them to add domain context, conventions, or playbooks.
A skill is a markdown file that gets concatenated onto an agent’s prompt. It’s the most direct way to teach an agent something about your codebase - your authentication model, the conventions you actually use, the playbook for a specific kind of change.
Where they live
Two roots are scanned:
.vibestrate/skills/- committed alongside your project. Use this for skills that should travel with the repo..claude/skills/- picked up when you’re already using Claude Code’s skill system locally.
Filename without the .md extension becomes the skill id. So auth-conventions.md is the skill auth-conventions.
A simple skill
# .vibestrate/skills/auth-conventions.md
This codebase uses Lucia for sessions. When touching auth:
- Don't create session middleware inline — use `requireSession` from `src/server/auth.ts`.
- Cookies are HttpOnly and SameSite=lax. Don't change those defaults.
- New auth routes go under `src/server/routes/auth/`.
That’s the whole format. Plain markdown, plain prose - agents read it the same way humans do.
Attach to an agent
In project.yml:
agents:
planner:
skills: [auth-conventions, error-handling]
executor:
skills: [auth-conventions]
Or per-run, with --skills:
vibe run "Add 2FA enrollment" --skills auth-conventions,security-review
Skills passed via --skills are merged with whatever’s configured on each agent - they don’t replace.
List what’s available
vibe skills list
vibe skills show auth-conventions
When to write a new skill
Write a skill when:
- You’re repeating the same context in multiple task prompts.
- The agent keeps making the same wrong assumption that you have to correct.
- There’s a convention that’s not documented anywhere else in the codebase.
Don’t write a skill when:
- The information already lives in CLAUDE.md or your project README - agents will read those naturally as part of project rules.
- It’s a one-off - just include it in the task description.
- It’s about a single file - comments in the file itself work better.
Reference
See the skill discovery and schema reference for the full filesystem layout and any optional metadata.