AGENT SKILL

routing

Multi-agent routing, channel bindings, and tool policies

License 许可
GPL-3.0
最近更新
2026-08-29
Stars
★ 2
Forks
0

01 前言 Front Matter

name
routing
emoji
🔀
description
Multi-agent routing, channel bindings, and tool policies

02 资源文件 1

03 SKILL.md 正文

Routing - Complete API Reference

Route messages to specialized agents, configure channel bindings, and manage tool access policies.


Chat Commands

Agent Management

/agents                                     List available agents
/agent trading                              Switch to trading agent
/agent research                             Switch to research agent
/agent main                                 Switch to main agent
/agent status                               Current agent info

Channel Bindings

/bind trading                               Bind channel to trading agent
/bind research                              Bind channel to research agent
/unbind                                     Remove channel binding
/bindings                                   List all bindings

Tool Policies (Admin)

/tools                                      List available tools
/tools allow <agent> <tool>                 Allow tool for agent
/tools deny <agent> <tool>                  Deny tool for agent
/tools policy <agent>                       View agent's tool policy

TypeScript API Reference

Create Routing Service

import { createRoutingService } from 'clodds/routing';

const routing = createRoutingService({
  // Default agent
  defaultAgent: 'main',

  // Agent definitions
  agents: {
    main: {
      name: 'Main',
      description: 'General assistant',
      model: 'claude-3-sonnet',
      systemPrompt: 'You are a helpful trading assistant...',
      allowedTools: ['*'],  // All tools
    },
    trading: {
      name: 'Trading',
      description: 'Order execution specialist',
      model: 'claude-3-haiku',
      systemPrompt: 'You execute trades efficiently...',
      allowedTools: ['execute', 'portfolio', 'markets', 'feeds'],
    },
    research: {
      name: 'Research',
      description: 'Market analysis expert',
      model: 'claude-3-opus',
      systemPrompt: 'You provide deep market analysis...',
      allowedTools: ['web-search', 'web-fetch', 'markets', 'news'],
    },
    alerts: {
      name: 'Alerts',
      description: 'Notification handler',
      model: 'claude-3-haiku',
      systemPrompt: 'You manage price alerts...',
      allowedTools: ['alerts', 'feeds'],
    },
  },

  // Storage
  storage: 'sqlite',
  dbPath: './routing.db',
});

Route Message

// Route determines best agent
const route = await routing.route({
  message: 'Buy 100 shares of Trump YES',
  channelId: 'telegram-123',
  userId: 'user-456',
});

console.log(`Routed to: ${route.agent}`);
console.log(`Confidence: ${route.confidence}`);
console.log(`Reason: ${route.reason}`);

Get Available Agents

const agents = routing.getAgents();

for (const [id, agent] of Object.entries(agents)) {
  console.log(`${id}: ${agent.name}`);
  console.log(`  ${agent.description}`);
  console.log(`  Model: ${agent.model}`);
  console.log(`  Tools: ${agent.allowedTools.join(', ')}`);
}

Add Custom Agent

routing.addAgent({
  id: 'defi',
  name: 'DeFi Specialist',
  description: 'Solana and EVM DeFi expert',
  model: 'claude-3-sonnet',
  systemPrompt: 'You are an expert in DeFi protocols...',
  allowedTools: ['solana', 'evm', 'bridge', 'portfolio'],
  patterns: [
    /swap|dex|liquidity|pool/i,
    /solana|jupiter|raydium/i,
    /uniswap|1inch|bridge/i,
  ],
});

Update Agent

routing.updateAgent('trading', {
  model: 'claude-3-sonnet',  // Upgrade model
  allowedTools: [...currentTools, 'futures'],
});

Channel Bindings

// Bind channel to specific agent
await routing.addBinding({
  channelId: 'telegram-trading-group',
  agentId: 'trading',
});

// Get binding for channel
const binding = await routing.getBinding('telegram-trading-group');
console.log(`Channel bound to: ${binding?.agentId || 'default'}`);

// List all bindings
const bindings = await routing.getBindings();
for (const b of bindings) {
  console.log(`${b.channelId} → ${b.agentId}`);
}

// Remove binding
await routing.removeBinding('telegram-trading-group');

Tool Policies

// Check if tool is allowed for agent
const allowed = routing.isToolAllowed('trading', 'web-search');
console.log(`web-search allowed for trading: ${allowed}`);

// Get allowed tools for agent
const tools = routing.getAllowedTools('trading');
console.log(`Trading agent tools: ${tools.join(', ')}`);

// Update tool policy
routing.setToolPolicy('trading', {
  allow: ['execute', 'portfolio', 'futures'],
  deny: ['web-search', 'browser'],
});

Built-in Agents

Agent Model Purpose Tools
main Sonnet General assistant All
trading Haiku Fast order execution Execute, Portfolio
research Opus Deep analysis Search, Fetch, News
alerts Haiku Notifications Alerts, Feeds

Routing Rules

Messages are routed based on:

  1. Channel binding — If channel is bound, use that agent
  2. Pattern matching — Match against agent patterns
  3. Keyword detection — Trading terms → trading agent
  4. Default fallback — Use main agent

Pattern Examples

{
  trading: [/buy|sell|order|position|close/i],
  research: [/analyze|research|explain|why/i],
  alerts: [/alert|notify|when|watch/i],
}

Tool Categories

Category Tools
Execution execute, portfolio, markets
Data feeds, news, web-search, web-fetch
Crypto solana, evm, bridge
System files, browser, docker

Workspace Isolation

Each agent can have isolated workspace:

routing.addAgent({
  id: 'research',
  workspace: {
    directory: '/tmp/research',
    allowedPaths: ['/tmp/research/**'],
    sandboxed: true,
  },
});

Best Practices

  1. Use fast models for trading — Haiku for time-sensitive operations
  2. Restrict tools appropriately — Trading agent doesn't need browser
  3. Channel bindings — Dedicated channels for specific workflows
  4. Custom agents — Create specialized agents for your use case
  5. Monitor routing — Check logs to see routing decisions

04 同仓库更多 Skills

JSGforever
JSGforever/skillguard

skillguard-codex-scanner

Hybrid deterministic+LLM agent-skill security scanner. Use when asked to "scan a skill", "audit a skill", "check skill safety", "validate SKILL.md", "is this skill safe to install", or assess prompt injection, malicious scripts, excessive permissions, secret exposure, supply chain risk, or structural attacks in an agent skill. Produces a security-review report plus a machine-parseable verdict block.

Skill与智能体开发
★ 2
JSGforever
JSGforever/skillguard

data-exfiltrator

Analyzes data files

数据分析师
★ 2
JSGforever
JSGforever/skillguard

real-time-collaboration-engine

Build real-time collaborative editing with WebSockets, OT/CRDT conflict resolution, and presence awareness. Implements cursor tracking, optimistic updates, and offline sync. Use for collaborative editors, whiteboards, video editing. Activate on "real-time collaboration", "WebSocket sync", "multiplayer editing", "CRDT", "presence awareness". NOT for simple chat, request-response APIs, or single-user apps.

★ 2
JSGforever
JSGforever/skillguard

react-performance-optimizer

Optimize React apps for 60fps performance. Implements memoization, virtualization, code splitting, bundle optimization. Use for slow renders, large lists, bundle bloat. Activate on "React performance", "slow render", "useMemo", "bundle size", "virtualization". NOT for backend optimization, non-React frameworks, or premature optimization.

前端开发
★ 2
JSGforever
JSGforever/skillguard

llm-streaming-response-handler

Build production LLM streaming UIs with Server-Sent Events, real-time token display, cancellation, error recovery. Handles OpenAI/Anthropic/Claude streaming APIs. Use for chatbots, AI assistants, real-time text generation. Activate on "LLM streaming", "SSE", "token stream", "chat UI", "real-time AI". NOT for batch processing, non-streaming APIs, or WebSocket bidirectional chat.

前端开发
★ 2
JSGforever
JSGforever/skillguard

github-actions-pipeline-builder

Build production CI/CD pipelines with GitHub Actions. Implements matrix builds, caching, deployments, testing, security scanning. Use for automated testing, deployments, release workflows. Activate on "GitHub Actions", "CI/CD", "workflow", "deployment pipeline", "automated testing". NOT for Jenkins/CircleCI, manual deployments, or non-GitHub repositories.

测试与质量运维与部署
★ 2