I wanted a way to combine agents by role instead of being locked into a single built-in workflow. In practice, I often want one agent acting as a manager and others acting as workers for implementation, review, or research.
The goal is to make that workflow usable from a normal terminal: start workers, delegate tasks, review progress, and coordinate everything from one control point.
Setup is currently:
uv tool install subagent-cli subagent config init --scope user npx skills add "github/otakumesi/subagent-cli"
Then you give your coding agent a task and have it coordinate worker agents via the `subagent-manager` skill.
The README includes a real Gemini → Codex demo.
The current scope is intentionally narrow: it’s alpha, local-only, single-host, and currently ACP-based.
One design question I’m still evaluating is the abstraction boundary. Right now, the CLI maps fairly directly to the protocol, and I’m not fully sure whether that’s the right tradeoff. The current worker / turn / approval / handoff model may be the right boundary, or some of it may need to be hidden behind a simpler interface.
I’d especially love feedback on whether this feels useful in real workflows, and whether the current model feels like the right UX for multi-agent coding.