agent-swarm
от runeweaverstudios
Web & Frontend DevelopmentКратко
IMPORTANT: OpenRouter is required.
Описание
Что делает этот навык
Routes incoming tasks to the appropriate AI model via OpenRouter, then delegates execution to a sub-agent through sessions_spawn. Tasks are classified into eight tiers, including code, reasoning, creative, and vision, with configurable primary and fallback models per tier. The orchestrator never answers tasks directly.
Instead of running every task on one expensive model, it classifies work into tiers and selects the cheapest capable model per task, cutting both cost and latency.
Когда его использовать
- Routing a code debugging request to a dedicated coding model
- Splitting a mixed task into parallel sub-agents for faster completion
- Sending screenshot analysis to a vision-capable model automatically
- Using a cheap fast model for simple status checks to reduce cost
- Delegating research queries to a search-optimized model
Пример сценария
User asks: fix the null pointer bug in auth.py and write release notes
- 1
Router receives the task and detects two distinct subtasks
- 2
Runs router.py with --multi flag to generate separate spawn configs for CODE and CREATIVE tiers
- 3
Calls sessions_spawn with the CODE-tier model for the bug fix
- 4
Calls sessions_spawn with the CREATIVE-tier model for the release notes
- 5
Returns combined results from both sub-agents to the user
Bug fix applied by code model, release notes written by creative model, both results returned together
Требования
- OpenRouter API key configured in OpenClaw platform settings
Установка
clawhub install runeweaverstudios/agent-swarmВыполните команду CLI или вставьте промпт в Claude Code, Cursor, Codex — любой кодинг-агент, которым пользуетесь.