Диспетчеризация задач между агентами в agentic-flow

orchestration

worker-integration is a Claude Code skill that provides intelligent task dispatch between background workers and specialized agents in agentic-flow. It maps eight trigger types — ultralearn, optimize, audit, benchmark, testgaps, document, deepdive, and refactor — to primary and fallback agents, each with a defined sequence of pipeline phases. Agent selection is driven by execution history: quality score, success rate, average latency, and execution count are combined to produce a confidence value (e.g., 0.87 after 45 runs at 94.2% success rate). Results are stored under a consistent `{trigger}/{topic}/{phase}` memory key pattern, and performance is monitored against configurable thresholds for p95 latency, memory usage, and scan coverage. Teams building multi-agent automation pipelines who need data-driven agent routing and continuous performance improvement will find this skill especially useful.