ruvnet__ruflo/v3/@claude-flow/plugins

@claude-flow/plugins

Unified Plugin SDK for Claude Flow V3

A comprehensive plugin development framework providing workers, hooks, providers, and security utilities for building Claude Flow extensions.

Installation

npm install @claude-flow/plugins

Quick Start

Create a Plugin with the Builder

import { PluginBuilder, HookEvent, HookPriority } from '@claude-flow/plugins';

const myPlugin = new PluginBuilder('my-awesome-plugin', '1.0.0')
  .withDescription('My awesome plugin for Claude Flow')
  .withAuthor('Your Name')
  .withMCPTools([
    {
      name: 'greet',
      description: 'Greet a user',
      inputSchema: {
        type: 'object',
        properties: {
          name: { type: 'string', description: 'Name to greet' }
        },
        required: ['name']
      },
      handler: async (input) => ({
        content: [{ type: 'text', text: `Hello, ${input.name}!` }]
      })
    }
  ])
  .withHooks([
    {
      event: HookEvent.PostTaskComplete,
      priority: HookPriority.Normal,
      handler: async (ctx) => {
        console.log('Task completed:', ctx.data);
        return { success: true };
      }
    }
  ])
  .build();

// Register with the default registry
import { getDefaultRegistry } from '@claude-flow/plugins';
await getDefaultRegistry().register(myPlugin);

Quick Plugin Creators

import { createToolPlugin, createHooksPlugin, createWorkerPlugin } from '@claude-flow/plugins';

// Tool-only plugin
const toolPlugin = createToolPlugin('my-tools', '1.0.0', [
  { name: 'tool1', description: '...', inputSchema: {...}, handler: async () => {...} }
]);

// Hooks-only plugin
const hooksPlugin = createHooksPlugin('my-hooks', '1.0.0', [
  { event: HookEvent.PreTaskExecute, handler: async (ctx) => ({ success: true }) }
]);

// Worker plugin
const workerPlugin = createWorkerPlugin('my-workers', '1.0.0', [
  { type: 'coder', name: 'main-coder', capabilities: ['code-generation'] }
]);

Features

🔧 MCP Tool Builder

import { MCPToolBuilder } from '@claude-flow/plugins';

const tool = new MCPToolBuilder('calculate')
  .withDescription('Perform calculations')
  .addStringParam('expression', 'Math expression', { required: true })
  .addBooleanParam('verbose', 'Show steps', { default: false })
  .withHandler(async (input) => {
    const result = eval(input.expression); // Use a safe evaluator in production!
    return { content: [{ type: 'text', text: `Result: ${result}` }] };
  })
  .build();

🎣 Hook System

import { HookBuilder, HookFactory, HookRegistry, HookEvent, HookPriority } from '@claude-flow/plugins';

// Create a custom hook with conditions
const hook = new HookBuilder(HookEvent.PreAgentSpawn)
  .withName('validate-agent')
  .withPriority(HookPriority.High)
  .when((ctx) => ctx.data?.type === 'coder')
  .transform((data) => ({ ...data, validated: true }))
  .handle(async (ctx) => {
    // Validation logic
    return { success: true, data: ctx.data, modified: true };
  })
  .build();

// Pre-built hook factories
const logger = HookFactory.createLogger(HookEvent.PostTaskComplete, console);
const rateLimiter = HookFactory.createRateLimiter(HookEvent.PreToolCall, { maxPerMinute: 100 });
const validator = HookFactory.createValidator(HookEvent.PreAgentSpawn, (data) => data.type !== undefined);

👷 Worker Pool

import { WorkerPool, WorkerFactory } from '@claude-flow/plugins';

// Create a worker pool
const pool = new WorkerPool({
  minWorkers: 2,
  maxWorkers: 10,
  taskQueueSize: 100
});

// Spawn workers using factory
const coder = await pool.spawn(WorkerFactory.createCoder('main-coder'));
const reviewer = await pool.spawn(WorkerFactory.createReviewer('code-reviewer'));
const tester = await pool.spawn(WorkerFactory.createTester('test-runner'));

// Submit tasks
const result = await pool.submit({
  id: 'task-1',
  type: 'code-generation',
  input: { prompt: 'Write a function...' }
});

// Shutdown
await pool.shutdown();

🤖 LLM Provider Integration

import { ProviderRegistry, ProviderFactory, BaseLLMProvider } from '@claude-flow/plugins';

const registry = new ProviderRegistry({
  fallbackChain: ['anthropic', 'openai'],
  costOptimization: true
});

// Register built-in providers
class ClaudeProvider extends BaseLLMProvider {
  constructor() {
    super(ProviderFactory.createClaude());
  }

  async complete(request) {
    // Implementation
  }
}

registry.register(new ClaudeProvider());

// Execute with automatic fallback
const response = await registry.execute({
  model: 'claude-sonnet-4-6',
  messages: [{ role: 'user', content: 'Hello!' }]
});

🔗 Agentic Flow Integration

import { AgenticFlowBridge, AgentDBBridge } from '@claude-flow/plugins';

// Swarm coordination
const agentic = new AgenticFlowBridge({ maxConcurrentAgents: 15 });
await agentic.initializeSwarm({ type: 'hierarchical', maxAgents: 15 });

const agent = await agentic.spawnAgent({
  type: 'coder',
  capabilities: ['typescript', 'react']
});

const result = await agentic.orchestrateTask({
  taskType: 'code-generation',
  input: { prompt: '...' },
  agentId: agent.id
});

// Vector storage with AgentDB
const agentdb = new AgentDBBridge({ dimensions: 1536, indexType: 'hnsw' });
await agentdb.initialize();

await agentdb.store('doc-1', embeddings, { type: 'document' });
const similar = await agentdb.search(queryVector, { limit: 10 });

🔒 Security Utilities

import { Security, createRateLimiter, createResourceLimiter } from '@claude-flow/plugins';

// Input validation
const name = Security.validateString(input, { minLength: 1, maxLength: 100 });
const count = Security.validateNumber(input, { min: 0, max: 1000, integer: true });
const path = Security.validatePath(input, { allowedExtensions: ['.ts', '.js'] });

// Safe path creation (prevents traversal attacks)
const safePath = Security.safePath('/project', 'src', userInput);

// Safe JSON parsing (prevents prototype pollution)
const data = Security.safeJsonParse<Config>(jsonString);

// Command validation
const cmd = Security.validateCommand('npm install', { allowedCommands: new Set(['npm', 'npx']) });

// Rate limiting
const limiter = createRateLimiter({ maxTokens: 100, refillRate: 10, refillInterval: 1000 });
if (limiter.tryAcquire()) {
  // Proceed
}

// Resource limiting
const resourceLimiter = createResourceLimiter({ maxMemoryMB: 512, maxExecutionTime: 30000 });
const result = await resourceLimiter.enforce(async () => {
  // Heavy computation
});

API Reference

Core Exports

Export	Description
PluginBuilder	Fluent builder for creating plugins
BasePlugin	Abstract base class for plugins
PluginRegistry	Plugin lifecycle management
getDefaultRegistry()	Get the default plugin registry

SDK Builders

Export	Description
MCPToolBuilder	Build MCP tools with parameters
HookBuilder	Build hooks with conditions and transformers
WorkerBuilder	Build worker definitions

Quick Creators

Export	Description
createToolPlugin()	Create a tool-only plugin
createHooksPlugin()	Create a hooks-only plugin
createWorkerPlugin()	Create a worker plugin
createProviderPlugin()	Create a provider plugin

Hook System

Export	Description
HookRegistry	Central hook management
HookExecutor	Execute hooks with patterns
HookFactory	Pre-built hook creators
HookEvent	All hook event types
HookPriority	Hook priority levels

Workers

Export	Description
WorkerPool	Managed worker pool
WorkerInstance	Individual worker
WorkerFactory	Worker definition factory

Providers

Export	Description
ProviderRegistry	LLM provider management
BaseLLMProvider	Base provider implementation
ProviderFactory	Provider definition factory

Integrations

Export	Description
AgenticFlowBridge	agentic-flow@alpha integration
AgentDBBridge	AgentDB vector storage

Security

Export	Description
Security	All security utilities
validateString/Number/Boolean/Array/Enum	Input validators
safePath/safePathAsync	Path security
safeJsonParse/safeJsonStringify	JSON security
createRateLimiter	Rate limiting
createResourceLimiter	Resource limiting

Hook Events

enum HookEvent {
  // Session lifecycle
  SessionStart = 'session:start',
  SessionEnd = 'session:end',

  // Agent lifecycle
  PreAgentSpawn = 'agent:pre-spawn',
  PostAgentSpawn = 'agent:post-spawn',
  PreAgentTerminate = 'agent:pre-terminate',
  PostAgentTerminate = 'agent:post-terminate',

  // Task lifecycle
  PreTaskExecute = 'task:pre-execute',
  PostTaskComplete = 'task:post-complete',
  TaskError = 'task:error',

  // Tool lifecycle
  PreToolCall = 'tool:pre-call',
  PostToolCall = 'tool:post-call',

  // Memory operations
  PreMemoryStore = 'memory:pre-store',
  PostMemoryStore = 'memory:post-store',
  PreMemoryRetrieve = 'memory:pre-retrieve',
  PostMemoryRetrieve = 'memory:post-retrieve',

  // Swarm coordination
  SwarmInitialized = 'swarm:initialized',
  SwarmShutdown = 'swarm:shutdown',
  ConsensusReached = 'swarm:consensus-reached',

  // File operations
  PreFileRead = 'file:pre-read',
  PostFileRead = 'file:post-read',
  PreFileWrite = 'file:pre-write',
  PostFileWrite = 'file:post-write',

  // Commands
  PreCommand = 'command:pre-execute',
  PostCommand = 'command:post-execute',

  // Learning
  PatternLearned = 'learning:pattern-learned',
  PatternApplied = 'learning:pattern-applied',
}

Hook Priorities

enum HookPriority {
  Critical = 1000,  // Run first, can abort
  High = 750,       // Important hooks
  Normal = 500,     // Default priority
  Low = 250,        // Less important
  Deferred = 0,     // Run last
}

Worker Types

• coder - Code implementation
• reviewer - Code review
• tester - Test generation/execution
• researcher - Information gathering
• planner - Task planning
• coordinator - Multi-agent coordination
• security - Security analysis
• performance - Performance optimization
• specialized - Custom capabilities
• long-running - Background tasks

Performance

Metric	Target	Actual
Plugin load time	< 50ms	~20ms
Hook execution	< 1ms	~0.5ms
Worker spawn	< 100ms	~50ms
Vector search (10K)	< 10ms	~5ms

Testing

npm test           # Run all tests
npm run test:watch # Watch mode

License

MIT