Skills
Reusable instructions vemlor injects into the agent's prompt.
A skill is a named instruction you write once and reuse across tasks. When a skill applies, vemlor prepends its body to the agent's prompt, so the agent follows your house rules without you retyping them every time.
Manage skills in Settings → Skills. They are workspace-scoped: any member can add or edit them, and every task in the workspace can use them.
Modes
Each skill has one of three modes:
- Off: disabled, never applied.
- On mention: applied only when you reference it with
/in the task prompt (for example/rails-conventions). - Always: applied automatically to every task in the workspace.
Switch modes any time. The mode a skill was in when a task ran is snapshotted onto that task, so changing it later never rewrites history.
Writing a skill
Give it a short name and a body of plain instructions. Keep bodies focused: one concern per skill composes better than one giant catch-all.
- Name: up to 80 characters.
- Body: up to 4000 characters.
Name: verify-before-pr
Body: Always run the test suite before opening a PR. If tests fail,
fix them or explain why in the PR description. Never push a red build.Example uses
- Coding conventions the agent should always follow.
- Project glossary or domain terms.
- A checklist to run before opening a pull request.
- Tone and formatting rules for PR descriptions.
For repo-specific guidance that lives in version control instead of your
workspace settings, see .vemlor.json.