Self-hosted, Self-Improving, Federated Everything App — dev management, agent orchestration, persistent mind, digital twin, second brain, and creative pipeline from a single dashboard over Tailscale to port the user into the OS
JavaScript ai-agentsdashboarddeveloper-toolsdigital-twinquantified-selfsecond-brainself-hostedtailscale
/**
* Per-agent debounced output batchers for the CoS Runner stream path.
*
* The runner emits `agent:output` per parsed line (see cos-runner/index.js), so
* a chatty agent would otherwise trigger a full state load+save per line. Each
* batcher coalesces a ~250ms window; we drain + drop it when the agent
* completes/errors so the final lines persist before the completion event.
*
* Extracted from `subAgentSpawner.js` (issue #2837) so `agentManagement.js` can
* import `flushRunnerOutputBatcher` statically. It previously reached for it via
* `await import('./subAgentSpawner.js')` purely to dodge a cycle — the spawner
* barrel re-exports agentManagement. Leaf module: only depends on the state
* layer (`cosAgentLifecycle.js`).
*/
import { createAgentOutputBatcher } from './cosAgentLifecycle.js';
const runnerOutputBatchers = new Map();
export function getRunnerOutputBatcher(agentId) {
let batcher = runnerOutputBatchers.get(agentId);
if (!batcher) {
batcher = createAgentOutputBatcher(agentId);
runnerOutputBatchers.set(agentId, batcher);
}
return batcher;
}
export async function flushRunnerOutputBatcher(agentId) {
const batcher = runnerOutputBatchers.get(agentId);
if (!batcher) return;
// Flush BEFORE deleting: the agent is still in `runnerAgents` at this point
// (handleAgentCompletion removes it afterwards), so a line racing in during
// the awaited flush lands in this same batcher instead of orphaning a new
// one. The `agent:output` guard in the spawner drops any truly
// post-completion stray.
await batcher.flush();
runnerOutputBatchers.delete(agentId);
}