# Complete Agent Integration: Prometheus + Every Code + Dexto ## Overview Integration of three major AI agent platforms into Claude Code CLI: 1. **Prometheus** (Python/LangGraph) - Multi-agent code reasoning with knowledge graphs 2. **Every Code** (Rust/Codex) - Fast terminal-based coding agent with Auto Drive 3. **Dexto** (TypeScript) - Agent harness/orchestration layer with MCP support ## Architecture Comparison | Aspect | Prometheus | Every Code | Dexto | |--------|-----------|------------|-------| | **Language** | Python | Rust | TypeScript | | **Framework** | LangGraph | Custom event-driven | Agent Harness | | **Core Role** | Code reasoning & bug fixing | Fast coding execution | Agent orchestration | | **Key Features** | Knowledge graphs (Neo4j), AST parsing, Docker | Auto Drive, Browser, Multi-agent | Session management, MCP, Tools | | **Agent Types** | Classification, Bug Reproduction, Context, Patch | Plan, Code, Solve, Auto Review | Custom agents via YAML | | **Code Understanding** | Neo4j + Tree-sitter AST | Native file system | Plugin-based | | **Execution** | Docker containers | Native shell | Tool orchestration | | **LLM Support** | OpenAI, Anthropic, Gemini | ChatGPT, Claude, Gemini | Any (via config) | | **MCP Support** | ❌ Native | ✅ Native | ✅ Native (client & server) | ## Unified Integration Architecture ``` ┌─────────────────────────────────────────────────────────────────────────┐ │ Claude Code CLI │ │ ┌────────────────────────────────────────────────────────────────────┐ │ │ │ Unified Agent Skills │ │ │ │ /agent-plan /agent-fix-bug /agent-solve │ │ │ │ /agent-context /agent-review /agent-orchestrate │ │ │ └─────────────────────────────────┬──────────────────────────────────┘ │ │ │ MCP Client │ └──────────────────────────────────┼──────────────────────────────────────┘ │ ┌──────────────────────────────────┼──────────────────────────────────────┐ │ Unified Agent MCP Server │ │ ┌────────────────────────────────────────────────────────────────────┐ │ │ │ Intelligent Router │ │ │ │ Routes to best backend for each task │ │ │ └─────┬────────────────┬────────────────┬────────────────────────────┘ │ │ │ │ │ │ │ ┌─────▼─────┐ ┌─────▼─────┐ ┌─────▼─────┐ │ │ │ Prometheus │ │Every Code │ │ Dexto │ │ │ │ │ │ │ │ │ │ │ │ • KG │ │ • Auto │ │ • Harness │ │ │ │ • AST │ │ Drive │ │ • MCP │ │ │ │ • Docker │ │ • Browser │ │ • Tools │ │ │ │ • LangGraph│ │ • Multi │ │ • Session │ │ │ │ │ │ Agent │ │ • YAML │ │ │ └───────────┘ └───────────┘ └───────────┘ │ └─────────────────────────────────────────────────────────────────────────┘ ``` ## Backend Specialization ### Prometheus - Deep Code Understanding **Best for:** - Complex bug fixing with reproduction - Semantic code search via AST - Large codebase understanding - Regression testing - Documentation analysis **Unique Capabilities:** - Neo4j knowledge graph with code relationships - Tree-sitter AST parsing for precise code analysis - Docker containerized testing - LangGraph state machine orchestration ### Every Code - Fast Execution **Best for:** - Rapid code generation - Auto Drive multi-agent tasks - Browser automation - Quick iterations - Terminal-based workflows **Unique Capabilities:** - Native Rust performance - Auto Review background quality checks - Browser CDP integration - Multi-CLI orchestration (Claude, Gemini, etc.) ### Dexto - Agent Orchestration **Best for:** - Complex multi-step workflows - Session management - Tool composition - Custom agent creation - MCP client/server operations **Unique Capabilities:** - YAML-based agent configuration - Built-in session management - Native MCP support (client & server) - Tool orchestration and composition - Memory and context management ## MCP Tool Specification ### Core Tools (Available from all backends) | Tool | Description | Backend Priority | |------|-------------|------------------| | `agent_find_file` | Find files by pattern | All | | `agent_search_code` | Search code by text | Prometheus (AST) > Dexto > Every Code | | `agent_read_file` | Read file with context | All | | `agent_create_file` | Create new file | All | | `agent_edit_file` | Edit existing file | All | | `agent_run_command` | Execute shell command | Every Code > Dexto > Prometheus (Docker) | ### Prometheus-Exclusive Tools | Tool | Description | |------|-------------| | `prometheus_classify_issue` | Classify GitHub issues (bug/feature/question/doc) | | `prometheus_fix_bug` | End-to-end bug fixing with reproduction | | `prometheus_get_context` | Semantic code context from knowledge graph | | `prometheus_search_ast` | Search by AST node type (function/class/variable) | | `prometheus_search_docs` | Search documentation with context | | `prometheus_run_tests` | Run tests in Docker container | | `prometheus_verify_patch` | Verify patch with regression tests | | `prometheus_get_callers` | Find all callers of a function | | `prometheus_get_callees` | Find all functions called by a function | ### Every Code-Exclusive Tools | Tool | Description | |------|-------------| | `everycode_auto_drive` | Orchestrate multi-agent automation | | `everycode_plan` | Generate implementation plan | | `everycode_solve` | Coordinate multiple CLI agents | | `everycode_auto_review` | Background code review | | `everycode_browser_goto` | Navigate browser to URL | | `everycode_browser_click` | Click element in browser | | `everycode_browser_screenshot` | Capture browser screenshot | | `everycode_browser_type` | Type text in browser | | `everycode_theme_set` | Set UI theme | ### Dexto-Exclusive Tools | Tool | Description | |------|-------------| | `dexto_create_agent` | Create custom agent from YAML | | `dexto_run_agent` | Run configured agent | | `dexto_list_sessions` | List all agent sessions | | `dexto_resume_session` | Resume previous session | | `dexto_mcp_connect` | Connect to MCP server | | `dexto_mcp_list_tools` | List available MCP tools | | `dexto_orchestrate` | Orchestrate multi-agent workflow | | `dexto_compose_tools` | Compose multiple tools into workflow | | `dexto_memory_store` | Store information in agent memory | | `dexto_memory_retrieve` | Retrieve from agent memory | ## Claude Code Skills ### Primary Skills ``` .claude/skills/agents/ ├── plan.md - Generate implementation plans ├── fix-bug.md - Fix bugs end-to-end ├── solve.md - Solve complex problems ├── review.md - Review code ├── context.md - Get code context ├── implement.md - Implement features ├── orchestrate.md - Orchestrate multi-agent workflows ├── test.md - Run and analyze tests └── README.md - Complete documentation ``` ### Skill Backend Selection | Skill | Primary | Secondary | Enhancement | |-------|---------|-----------|-------------| | `/agent-plan` | Every Code | Dexto | Prometheus (context) | | `/agent-fix-bug` | Prometheus | Dexto | Every Code (review) | | `/agent-solve` | Every Code | Dexto | Prometheus (analysis) | | `/agent-review` | Every Code | Dexto | Prometheus (AST) | | `/agent-context` | Prometheus | Dexto | - | | `/agent-implement` | Every Code | Dexto | Prometheus (validation) | | `/agent-orchestrate` | Dexto | Every Code | Prometheus | | `/agent-test` | Prometheus | Dexto | Every Code | ## Installation Guide ### Prerequisites ```bash # System requirements - Python 3.11+ - Rust + Cargo - Node.js 20+ - TypeScript - Docker + Docker Compose - Neo4j (for Prometheus knowledge graph) # Clone repositories git clone https://github.com/EuniAI/Prometheus.git ~/Prometheus git clone https://github.com/just-every/code.git ~/EveryCode git clone https://github.com/truffle-ai/dexto.git ~/Dexto ``` ### Prometheus Setup ```bash cd ~/Prometheus pip install -r requirements.txt cp example.env .env # Edit .env with API keys (OpenAI, Anthropic, or Gemini) # Start Neo4j docker-compose up -d neo4j # Build knowledge graph for your repo python -m prometheus.script.build_kg --repo_path /path/to/repo ``` ### Every Code Setup ```bash cd ~/EveryCode npm install -g @just-every/code # Authenticate code # Choose "Sign in with ChatGPT" or set OPENAI_API_KEY # Test installation code --version ``` ### Dexto Setup ```bash cd ~/Dexto npm install npm run build # Configure cp config.example.yaml config.yaml # Edit config.yaml with your settings # Test npm run start ``` ### Unified MCP Server Installation ```bash # Install all three MCP servers pip install git+https://github.com/your-org/prometheus-mcp.git pip install git+https://github.com/your-org/everycode-mcp.git npm install -g @your-org/dexto-mcp # Configure Claude Code # Add to ~/.config/claude-code/config.json: { "mcpServers": { "prometheus": { "command": "prometheus-mcp", "args": ["--repo", "/path/to/your/repo"], "env": { "OPENAI_API_KEY": "sk-...", "NEO4J_URI": "bolt://localhost:7687" } }, "everycode": { "command": "everycode-mcp", "args": ["--repo", "/path/to/your/repo"] }, "dexto": { "command": "dexto-mcp", "args": ["--config", "~/Dexto/config.yaml"] } } } ``` ## Usage Examples ### Example 1: Bug Fixing ``` /agent-fix-bug "Login fails after password reset" # Backend orchestration: # 1. Prometheus: Classify issue → Bug # 2. Prometheus: Reproduce in Docker # 3. Prometheus: Generate patch via LangGraph # 4. Dexto: Orchestrate verification workflow # 5. Every Code: Auto Review validates quality # 6. Prometheus: Run regression tests ``` ### Example 2: Feature Implementation ``` /agent-plan "Add real-time notifications with WebSocket" /agent-implement # Backend orchestration: # 1. Every Code: Generate implementation plan # 2. Prometheus: Get context from similar features # 3. Dexto: Create custom agent for implementation # 4. Every Code: Execute code generation # 5. Prometheus: Verify with AST analysis ``` ### Example 3: Complex Orchestration ``` /agent-orchestrate "Migrate auth system to OAuth2 Steps: audit current system, design OAuth2 flow, implement, test, deploy" # Backend orchestration: # 1. Dexto: Create multi-agent workflow # 2. Prometheus: Audit current auth (AST + KG) # 3. Every Code: Design OAuth2 flow # 4. Dexto: Coordinate implementation agents # 5. Prometheus: Run integration tests # 6. Dexto: Manage deployment pipeline ``` ## Success Metrics 1. ✅ All three platforms accessible via MCP 2. ✅ Intelligent backend routing working 3. ✅ All 8 core skills functional 4. ✅ End-to-end workflows working 5. ✅ Multi-platform orchestration working 6. ✅ Documentation complete 7. ✅ Setup scripts available ## Implementation Timeline ### Phase 1: Foundation (Week 1) - [x] Analyze all three architectures - [x] Design unified integration strategy - [ ] Create MCP server skeletons - [ ] Implement intelligent router ### Phase 2: Platform Integrations (Week 2-3) - [ ] Prometheus MCP server complete - [ ] Every Code MCP server complete - [ ] Dexto MCP server complete - [ ] Test each platform independently ### Phase 3: Claude Code Skills (Week 3-4) - [ ] Implement all 8 core skills - [ ] Add backend selection logic - [ ] Create skill compositions - [ ] Add error handling ### Phase 4: Polish & Documentation (Week 4) - [ ] Performance optimization - [ ] Setup automation scripts - [ ] User documentation - [ ] Developer guide - [ ] Release ## Next Steps 1. ✅ Architectures analyzed (COMPLETE) 2. ⏳ Complete MCP server implementations 3. ⏳ Create comprehensive skill suite 4. ⏳ Write setup automation 5. ⏳ Test all integrations 6. ⏳ Document and release