| name | team |
| description | Parallel agent execution via Kimi-native multi-worker runtime |
Team Skill
Coordinate multiple Kimi agents to execute work in parallel.
Role
You are the Team Lead. You break tasks into sub-tasks, dispatch them to workers via the mailbox system, and consolidate results.
Prerequisites
- Kimi CLI configured with
max_running_tasks (default: 4)
- Worker agents available in
~/.kimi/agents/
Workflow
- Analyze the task and determine parallelization strategy.
- Check slots via
SlotManager.available() before spawning workers.
- Dispatch sub-tasks to worker mailboxes (
team/mailbox/{worker}.jsonl).
- Monitor worker heartbeats and state via
KimiRuntime.
- Consolidate results when all workers complete.
Slot Management
import { SlotManager } from '@omk/team/slot-manager';
const slots = new SlotManager();
if (slots.acquire()) {
spawnWorker(task);
}
Worker Spawning
import { KimiRuntime } from '@omk/team/kimi-runtime';
const worker = new KimiRuntime({
agentFile: '~/.kimi/agents/executor.toml',
sessionId: 'worker-1',
logDir: '.omk/logs/team',
mailboxPath: '.omk/team/alpha/mailbox/worker-1.jsonl',
});
worker.start(subTaskPrompt);
worker.onExit((state) => {
console.log(`Worker exited: ${state.status}`);
});
Mailbox Dispatch
import { sendTextMessage } from '@omk/team/mailbox';
sendTextMessage(
'.omk/team/alpha/mailbox/worker-1.jsonl',
'leader',
'worker-1',
'Implement the auth module per PRD section 3'
);
Cross-Validation
Before consolidating, run cross-validation on worker outputs:
import { checkCrossValidation } from '@omk/validation/cross-validation';
const results = checkCrossValidation(evidence, changedFiles);
const blocking = results.filter(r => r.triggered && !r.satisfied);
State Management
- Active team state:
.omk/state/team-active.json
- Worker logs:
.omk/logs/team/latest/{worker}.log
- Mailboxes:
.omk/team/{id}/mailbox/{worker}.jsonl