DevDiff

Appearance

DevDiff Agent Orchestration & OpenClaw Integration Platform ​

The DevDiff Agent Orchestration Platform (introduced in DevDiff v1.7.0) establishes DevDiff as the shared memory, execution engine, and coordination layer for multi-agent swarms (OpenClaw, Copilot, Gemini, Claude, and custom agents).

🎯 Architecture Vision ​

┌─────────────────────────────────────────────────────────────┐
│              DEVDIFF — THE AGENTIC PLATFORM                   │
│                                                              │
│  "DevDiff isn't a tool agents can use.                       │
│   DevDiff is the platform agents LIVE on."                   │
│                                                              │
│  Every agent — OpenClaw, Copilot, Gemini, Claude, custom —   │
│  connects to DevDiff as their shared memory, execution        │
│  engine, and coordination layer.                              │
│                                                              │
│  Agents don't call DevDiff. Agents run THROUGH DevDiff.      │
└─────────────────────────────────────────────────────────────┘

🦞 OpenClaw Supervisor v2 (OpenClawSupervisorV2) ​

The OpenClawSupervisorV2 orchestrates specialized agent squads according to project configuration in .devdiff/agents/openclaw/supervisor.yaml:

1. Specialized Agent Roles ​

  • 🏗️ Architect: architecture_analysis, design_pattern_detection, refactoring_identification, dependency_mapping.
  • 🔒 Security: vulnerability_scanning, secret_detection, cve_matching, compliance_verification.
  • ⚡ Performance: complexity_analysis, memory_profiling, bundle_impact, query_optimization.
  • 📝 Documentation: changelog_generation, api_documentation, migration_guide, release_notes.
  • 💬 Q&A: code_explanation, architecture_questions, historical_context.

2. Task Graph Decomposition ​

The supervisor automatically decomposes user objectives into parallel and sequential subtasks:

  • changelog_generation: Parallel diff & context scanning ➔ Sequential architecture, security, and performance analysis ➔ Final documentation synthesis & quality review.
  • security_audit: Parallel secret, dependency, and injection scans ➔ Sequential code pattern & compliance checks ➔ Final report generation.
  • answer_question: Parallel memory & index lookups ➔ Sequential Q&A synthesis.

3. Automated Validation Gates ​

  • Auto-Approve Threshold: $\ge 85%$ confidence.
  • Human Review Threshold: $70% - 84%$ confidence.
  • Auto-Retry Threshold: $< 50%$ confidence.

🤖 Universal Agent Registry (AgentRegistry) ​

The AgentRegistry allows connecting ANY AI agent via OpenClaw, MCP, REST, or WebSockets:

typescript
import { AgentRegistry } from "@eldrex/core";

// Register custom agent
AgentRegistry.register({
  id: "custom-reviewer",
  name: "Custom Code Reviewer",
  type: "custom",
  protocol: "mcp",
  status: "active",
  capabilities: ["code_review"],
  currentLoad: 0.1,
  successRate: 95.0,
  execute: async (task) => ({ confidence: 95, findings: ["Approved"] }),
});

// Delegate task with fallback routing
const result = await AgentRegistry.delegate({
  id: "task-1",
  type: "security_audit",
  requiredCapabilities: ["secret_detection"],
  priority: "high",
  data: {},
});

// Run multi-agent swarm with pairwise consensus
const swarm = await AgentRegistry.swarm(
  {
    id: "task-2",
    type: "changelog_generation",
    requiredCapabilities: [],
    priority: "high",
    data: {},
  },
  3,
);

💻 CLI Commands ​

bash
# Deploy full multi-agent swarm on an objective
devdiff agent swarm "Analyze security and performance impact of new auth endpoints"

# Deploy specific agent roles
devdiff agent deploy --agents architect,security "Review PR changes"

# Query a specific agent role directly
devdiff agent ask architect --prompt "Explain the rate limiter pattern"

# Run parallel subtasks
devdiff agent parallel --tasks "security scan, performance analysis"

# Trigger inter-agent conversation bus
devdiff agent converse --agents architect,security "Caching Strategy Trade-offs"

# Render ASCII Swarm Dashboard
devdiff agent dashboard

Learn more on the official website: https://devdiff.vercel.app/