⚠️ WORKFLOW INITIALIZATION

If starting a new session, first read workflow initialization:

mcp__plugin_swe_serena__read_memory("wf/WF_INIT")

Follow WF_INIT instructions before executing this skill.

---

Swarm Analyze Skill

Deep codebase analysis using Decentralized Autonomous Agents (DAA).

When to Use

• Large codebases (1000+ files)
• Complex multi-module projects
• When detailed DOM_* and SYS_* memories are needed
• Feature onboarding with full analysis mode

MCP Requirements

Required (one of):

• ruv-swarm MCP (preferred for DAA learning)
• claude-flow MCP (alternative)

Fallback: Sequential analysis if no swarm MCP available

Agent Types

Agent ID	Purpose	Cognitive Pattern
config-analyzer	Parse config files	convergent
architecture-mapper	Detect layers	systems
pattern-detector	Find conventions	lateral
domain-extractor	Extract domains	divergent
system-finder	Identify systems	systems
test-analyzer	Test patterns	critical
import-tracer	Dependency graph	convergent
convention-learner	Style detection	adaptive
file-indexer	File inventory	convergent
synthesizer	Compile results	systems

Process

Step 1: Initialize Swarm

⚠️ CRITICAL: RUV-Swarm has TWO separate agent pools - choose ONE pattern:

Pattern	Agent Creation	Execution	Use When
Swarm	agent_spawn	task_orchestrate	Parallel task execution
DAA	daa_agent_create	daa_workflow_execute	Learning/adaptation needed

// Option A: RUV-Swarm Task Orchestration (faster, no learning)
if (mcp_available("ruv-swarm") && !needsLearning) {
  mcp__ruv-swarm__swarm_init({ topology: "mesh", strategy: "balanced", maxAgents: 10 });
}

// Option B: RUV-Swarm DAA Workflow (slower, with learning)
if (mcp_available("ruv-swarm") && needsLearning) {
  mcp__ruv-swarm__daa_init({ enableLearning: true, enableCoordination: true });
}

// Option C: Claude-Flow (alternative)
if (mcp_available("claude-flow")) {
  mcp__claude-flow__swarm_init({ topology: "mesh", maxAgents: 10 });
}

Step 2: Spawn Analysis Agents

CRITICAL: Spawn ALL agents in ONE message for parallelism

Option A: Swarm Agents (for task_orchestrate)

// These go into the SWARM pool - usable by task_orchestrate
mcp__ruv-swarm__agent_spawn({ type: "analyst", name: "config-analyzer" })
mcp__ruv-swarm__agent_spawn({ type: "analyst", name: "architecture-mapper" })
mcp__ruv-swarm__agent_spawn({ type: "researcher", name: "pattern-detector" })
mcp__ruv-swarm__agent_spawn({ type: "researcher", name: "domain-extractor" })
mcp__ruv-swarm__agent_spawn({ type: "analyst", name: "system-finder" })
mcp__ruv-swarm__agent_spawn({ type: "analyst", name: "test-analyzer" })
mcp__ruv-swarm__agent_spawn({ type: "researcher", name: "import-tracer" })
mcp__ruv-swarm__agent_spawn({ type: "researcher", name: "convention-learner" })
mcp__ruv-swarm__agent_spawn({ type: "analyst", name: "file-indexer" })
mcp__ruv-swarm__agent_spawn({ type: "coordinator", name: "synthesizer" })

Option B: DAA Agents (for daa_workflow_execute)

// These go into the DAA pool - usable by daa_workflow_execute, NOT task_orchestrate
const agents = [
  { id: "config-analyzer", cognitivePattern: "convergent" },
  { id: "architecture-mapper", cognitivePattern: "systems" },
  { id: "pattern-detector", cognitivePattern: "lateral" },
  { id: "domain-extractor", cognitivePattern: "divergent" },
  { id: "system-finder", cognitivePattern: "systems" },
  { id: "test-analyzer", cognitivePattern: "critical" },
  { id: "import-tracer", cognitivePattern: "convergent" },
  { id: "convention-learner", cognitivePattern: "adaptive" },
  { id: "file-indexer", cognitivePattern: "convergent" },
  { id: "synthesizer", cognitivePattern: "systems" }
];

// Spawn all DAA agents in parallel
agents.forEach(a => mcp__ruv-swarm__daa_agent_create({
  id: a.id,
  cognitivePattern: a.cognitivePattern,
  enableMemory: true,
  learningRate: 0.8
}));

Step 3: Orchestrate Analysis

⚠️ Match execution to agent type!

Option A: Swarm Agents → task_orchestrate

// ONLY works with agents from agent_spawn
mcp__ruv-swarm__task_orchestrate({
  task: "Analyze codebase structure, patterns, domains, and systems",
  strategy: "parallel",
  maxAgents: 10,
  priority: "high"
});

Option B: DAA Agents → daa_workflow_execute

// ONLY works with agents from daa_agent_create
mcp__ruv-swarm__daa_workflow_create({
  id: "analysis-workflow",
  name: "Codebase Analysis",
  strategy: "parallel"
});

mcp__ruv-swarm__daa_workflow_execute({
  workflowId: "analysis-workflow",
  agentIds: ["config-analyzer", "architecture-mapper", "pattern-detector",
             "domain-extractor", "system-finder", "test-analyzer",
             "import-tracer", "convention-learner", "file-indexer", "synthesizer"],
  parallelExecution: true
});

Step 4: Collect Results

Each agent produces structured findings:

• config-analyzer: package.json, framework configs
• architecture-mapper: layers, directories, data flow
• pattern-detector: naming conventions, import patterns
• domain-extractor: business domains, entities
• system-finder: external integrations, APIs
• test-analyzer: test framework, coverage patterns
• import-tracer: dependency graph
• convention-learner: code style, formatting
• file-indexer: file inventory by type
• synthesizer: combined analysis

Step 5: Generate Memories

Based on synthesized results, create:

1. FEATURE_[KEY] - Main feature memory
2. DOM_[KEY]_[domain] - For each detected domain
3. SYS_[KEY]_[system] - For each detected system
4. Update INDEX_FEATURES - Add feature entry
5. Update ARCH_INDEX - Add architecture details

Step 6: DAA Learning

Record analysis success for future improvement:

mcp__ruv-swarm__daa_agent_adapt({
  agentId: "synthesizer",
  performanceScore: 0.9,
  feedback: "Analysis complete"
});

mcp__ruv-swarm__daa_knowledge_share({
  sourceAgentId: "synthesizer",
  targetAgentIds: ["config-analyzer", "architecture-mapper"],
  knowledgeDomain: "codebase-patterns"
});

Output Format

SWARM ANALYSIS COMPLETE

Metric	Value
Agents Used	10
Analysis Time	[duration]

Detected:

• Language: [primary]
• Framework: [name]
• Layers: [count]
• Domains: [count]
• Systems: [count]

Memories Created:

• FEATURE_[KEY]
• DOM_[KEY]_[domain1]
• DOM_[KEY]_[domain2]
• SYS_[KEY]_[system1]
• INDEX_FEATURES (updated)
• ARCH_INDEX (updated)

DAA Learning:

• Patterns stored: [count]
• Confidence: [score]

Skill Return Format

## Skill Return

- **Skill**: swe-swarm-analyze
- **Status**: [success|success_with_findings|blocked]
- **Agents Used**: [count]
- **Memories Created**: [list]
- **Domains Found**: [count]
- **Systems Found**: [count]
- **Next Step Hint**: WF_CLASSIFY

Fallback: Sequential Analysis

If no swarm MCP available:

⚠️ No swarm MCP detected. Running sequential analysis.

This will take longer but produce similar results.

Progress:
[1/10] Analyzing config files...
[2/10] Mapping architecture...
...

Exit

> **Skill /swe-swarm-analyze complete** - [count] memories created via DAA analysis