Manage Subagents¶
Aikito keeps canonical subagent definitions in subagents.toml and the
subagents/ directory, then renders them into formats supported by each Agent.
Preview¶
Review the canonical definition and preview the rendering plan:
The plan identifies creates, updates, unsupported capabilities, or unmanaged target conflicts.
Apply and Verify¶
aikito sync subagents
aikito show subagents
aikito show subagent verifier
aikito edit subagent verifier
Managed files are updated from the canonical definition. An existing file without an Aikito marker is treated as unmanaged and will not be overwritten by default.
Example aikito show subagents output from a configured workspace:
┌───────────┬───────┬─────────────┬─────────────────┬──────────┬────────────────────┬──────────────────┐
│ Subagent │ Codex │ Claude Code │ Antigravity CLI │ OpenCode │ GitHub Copilot CLI │ DeepSeek Harness │
├───────────┼───────┼─────────────┼─────────────────┼──────────┼────────────────────┼──────────────────┤
│ verifier │ ✓ │ ✓ │ ✓ │ ✓ │ ✓ │ ✓ │
└───────────┴───────┴─────────────┴─────────────────┴──────────┴────────────────────┴──────────────────┘
✓ means the rendered definition is synchronized; – means that Agent does
not participate in subagent synchronization.
- Antigravity CLI: Definitions are rendered to
~/.gemini/config/agents/<name>/agent.md, supportingmodelandtools. - OpenCode: Definitions are rendered to
~/.config/opencode/agents/<name>.mdwithmode: subagent; set the native model ID through[subagents.<name>.opencode].model. - Pi: Participates only when its optional extension entry point exists at
~/.pi/agent/extensions/subagent/index.ts. Definitions are rendered to~/.pi/agent/agents/<name>.md. Pi-specific configuration supportsmodelandtools; without the extension, Pi remains skipped and Aikito writes nothing.
If a definition is removed, status may report a managed orphan. Review it
before using the command's explicit pruning or force options. Use
aikito sync subagents --help for the options supported by the installed
version.
See Architecture for Agent capability boundaries and Safety model before forcing any target.