AgentSkillsCN

swe-swarm-orchestrate

针对大型任务的多智能体蜂群协作。

SKILL.md
--- frontmatter
name: swe-swarm-orchestrate
version: 1.0.0
description: Multi-agent swarm coordination for large tasks
workflow:
  aware: true
  callable_from:
    - WF_SWARM_ORCHESTRATE
  default_return: WF_EXECUTE
  supports_standalone: false
  auto_transition: true

⚠️ WORKFLOW INITIALIZATION

If starting a new session, first read workflow initialization:

code
mcp__plugin_swe_serena__read_memory("WF_INIT")

Follow WF_INIT instructions before executing this skill.


Swarm Orchestrate Skill

Coordinate multi-agent swarm for complex tasks.

MCP Selection Priority

  1. Claude Flow (preferred) - General orchestration
  2. RUV-Swarm (fallback) - DAA learning agents
  3. Sequential (no MCP) - Chunked execution

Agent Types

AgentPurpose
researcherExplore codebase
coderImplement changes
analystReview patterns
optimizerPerformance tuning
coordinatorOrchestrate tasks

Swarm Topologies

  • mesh - All agents connected (default)
  • hierarchical - Tree structure
  • ring - Circular communication
  • star - Central coordinator

Actions

  1. Select swarm system based on available MCPs
  2. Decompose task into parallel subtasks
  3. Spawn agents (ALL in one message)
  4. Coordinate execution
  5. Synthesize results

Critical Rules

  • NEVER run npx claude-flow init - use MCP tools only
  • Spawn agents in parallel (single message)
  • Store state in WORKING_MEMORY

Exit

> **Skill /swe-swarm-orchestrate complete** - returning to WF_EXECUTE