monitor-loop

What problem does it solve?

This Skill prevents the orchestrator from blocking on full TaskOutput dumps and from being overwhelmed by verbose sub-agent transcripts by converting milestone signals into concise NDJSON events consumed by a Monitor.

Core Features & Use Cases

• Background execution: Spawn specialist sub-agents with run_in_background so the orchestrator proceeds without waiting on full transcripts.
• Event-driven coordination: Sub-agents append single-line NDJSON events to .cortex/events.log to signal task_done, contract_ready, milestone_reached, or error.
• Efficient monitoring: Use a Monitor that tails the event log with line-buffered grep to emit discrete notifications into the orchestrator's flow.
• Safety and recovery: Includes a timeout-based safety net to pull an individual agent's TaskOutput when events are missing and guidance on limiting Monitor volume.
• Use case: Orchestrate parallel waves of specialists (e.g., builders, testers, documenters) and advance waves only when authoritative task_done and contract_ready events arrive.

Quick Start

Initialise the .cortex/events.log, spawn specialist tasks with run_in_background enabled, and start a Monitor to tail the log for task_done and contract_ready events.