Agent Skills¶
VOR Stream bundles task-specific agent skills directly into the vor binary.
A skill is a short Markdown guide that teaches an AI coding agent (such as Claude
Code, Codex, or GitHub Copilot) how to accomplish a VOR Stream task with the
vor CLI. Because the skills ship inside the binary, the guidance an agent reads
always matches the installed version and never goes stale.
Two commands cover the workflow: one to read skills, and one to make your agent aware of them.
Discover and read skills¶
List the skills bundled in your vor build:
vor show skills
Print one skill's full content:
vor show skill <name>
When the output is piped or redirected (as when an agent runs it), the raw
Markdown is emitted unchanged so the agent receives clean text. When you run it
in a terminal, it is rendered for readability. Add --json to vor show skills
for a machine-readable catalog (this flag applies to the list only).
Enable skills for your agent¶
vor enable skills installs a single pointer skill into your AI agent's skill
directory. The pointer teaches the agent to run vor show skills to discover the
bundled guides and vor show skill <name> to load the relevant one.
With no flag it writes both skill directories, so every agent is covered:
vor enable skills
Use --agent (repeatable, or comma-separated) to scope it:
vor enable skills --agent claude-code # only Claude Code
vor enable skills --agent universal # only the shared .agents/skills directory
vor enable skills --agent '*' # all (same as no flag)
There are two skill directories, because that is what the agent ecosystem uses:
--agent value |
Skill directory | Used by |
|---|---|---|
claude-code |
.claude/skills |
Claude Code |
universal |
.agents/skills |
Codex, GitHub Copilot, Cursor, Cline, Gemini, and most other agents |
Common agent names (codex, copilot, cursor, cline, gemini, …) are
accepted as aliases for universal, so vor enable skills --agent codex works
as expected.
The command writes into the enclosing Git repository, is safe to re-run (it rewrites the pointer only when its content changes), and writes each target directory once even when several agents resolve to the same one.