Sub-agents

Sub-agents (also called custom agents, modes, or personas depending on the product) store reusable instructions and metadata that specialized agent runs can load.

Supported format

Library sub-agents include:

• Name and description
• Instruction content
• Optional subdirectory/metadata and tags

Adapters map this shared model onto native files (markdown agents, JSON custom agents, custom modes, and so on). Gated entries mean the native concept is related but not identical.

Create or import

1. Open Library → Sub Agents → New.
2. Write a clear description: many agents surface description text in pickers.
3. Save and assign to destinations/agents that support sub-agents.
4. Import existing native agents through sync review when detected.

Compatibility highlights

• Broad support: Claude Code, Codex, Cursor, Kiro, OpenCode, Roo Code, Augment, GitHub Copilot CLI, and others.
• Gated examples: GitHub Copilot, Kilo Code custom modes (semantic mapping warnings may appear).
• Unsupported examples: Aider, Antigravity file-backed reusable sync, Cline, CodeWhale, Devin, Grok, Hermes, Zed.

Antigravity may document runtime/CLI subagents while still leaving Visu file-backed reusable sync unsupported: use the matrix, not vendor marketing alone.

Linking and synchronization

• Sub-agents are assignable and synchronizable.
• Writes preserve unknown native fields where adapters implement merge behavior (important for Kiro JSON agents and similar formats).
• Prefer Library edits over dual editing native files.

Limitations

• Not every native agent option is exposed in the Visu form.
• Gated mode mappings may not round-trip every native field the same way a first-class agent file would.
• Unsupported agents will not receive assignments even if the Library item exists.