First — huge thanks for the work on Kandev. The orchestration model, the TUI, and the agent integrations are excellent, and the roadmap transparency is appreciated.
We noticed "Coordinator mode" is already on the Next bucket of the roadmap — this issue is a suggestion to scope it slightly more ambitiously.
Proposal
A persistent Conductor agent session that monitors and orchestrates other sessions:
- Watches all active task sessions and their state
- Auto-responds when confidence is high (routine "continue?" prompts, lint fixes, retries on known-flaky steps)
- Escalates to the human when it isn't confident
- Exposes Kandev primitives (create task, spawn agent, open PR, run workflow, query session state, post comment) as tool calls available to the conductor
Why
- Removes the human polling loop; one operator can supervise many parallel agents.
- Natural superset of the planned Coordinator mode — sub-task delegation is one of its tools, not its only purpose.
- Pairs well with the webhook/event idea in a sibling issue: external events feed the conductor, which decides what to spawn.
First — huge thanks for the work on Kandev. The orchestration model, the TUI, and the agent integrations are excellent, and the roadmap transparency is appreciated.
We noticed "Coordinator mode" is already on the Next bucket of the roadmap — this issue is a suggestion to scope it slightly more ambitiously.
Proposal
A persistent Conductor agent session that monitors and orchestrates other sessions:
Why