Skills

What Skills Do

Skills are reusable instruction packs. They can add workflows, constraints, examples, and MCP tool access without permanently bloating every agent prompt.

Category + Skill Pattern

Categories choose the model/profile. Skills provide the domain playbook.

task({
  category: "writing",
  load_skills: ["sync-openagent-docs"],
  run_in_background: false,
  prompt: "Refresh docs from upstream README and docs/reference facts.",
});

Rules of thumb:

• Use visual-engineering for UI work, even if the edit looks small.
• Load every skill whose domain overlaps the task.
• Prefer project/user skills over generic built-ins when both apply.
• Keep skill prompts concrete: context, goal, downstream use, request.

Documentation Sync Skill

This docs site includes sync-openagent-docs, a project skill for keeping content aligned with upstream code-yeongyu/oh-my-openagent.

It tells future agents to:

• Read upstream README, docs, package metadata, and relevant .opencode/skills files before editing.
• Preserve the package-name transition: install docs should mention oh-my-opencode while OpenAgent remains the product identity.
• Reduce diagram usage unless a diagram clarifies an architecture boundary.
• Update English and Korean first, then keep Japanese and Chinese routes navigable.
• Verify with lint, build, and route smoke checks.

Writing Good Project Skills