Skip to content

CLAUDE.md

Latest commit

 

History

History
47 lines (35 loc) · 1.9 KB

CLAUDE.md

File metadata and controls

47 lines (35 loc) · 1.9 KB

agentic-control-plane

Overview

A multi-agent demo showcasing AI agents as first-class Kubernetes citizens that collaborate to manage, monitor, and troubleshoot cluster workloads. Agents communicate via the A2A protocol, are discovered through AgentCard CRDs, and are bridged to supervisors via MCP.

Repository Structure

agentic-control-plane/
├── agents/
│   ├── k8s_debug_agent/        # AG2-based K8s debugging agent (A2A)
│   └── source_code_analyzer/   # AG2-based source code analysis agent (A2A)
├── tools/
│   ├── a2a_bridge_server/      # MCP-to-A2A bridge (agent discovery + invocation)
│   └── k8s_readonly_server/    # K8s read-only MCP tool server
└── deploy/                     # Kubernetes manifests (kustomize)

Key Commands

Task Command
Lint make lint
Format make fmt
Install pre-commit pre-commit install
Run agent (k8s-debug) cd agents/k8s_debug_agent && uv run python a2a_agent.py
Run agent (src-analyzer) cd agents/source_code_analyzer && uv run python a2a_agent.py
Run bridge cd tools/a2a_bridge_server && uv run python server.py
Deploy to K8s kubectl apply -k deploy/<component>/

Code Style

  • Python 3.10+, uv package manager
  • ruff for linting and formatting (line-length: 100)
  • Pre-commit hooks: pre-commit install
  • DCO sign-off required: git commit -s

Protocols

  • A2A: Agent-to-Agent (Google) — agents expose /.well-known/agent.json
  • MCP: Model Context Protocol — bridge translates MCP ↔ A2A
  • AgentCard CRD: Kubernetes CR caching agent capabilities for discovery

Architecture

The supervisor (Claude Code or in-cluster agent) discovers running agents via the A2A bridge, which reads AgentCard CRDs created by the kagenti-operator. Agents expose skills that the supervisor delegates to for K8s operations and code analysis.