Agent Capabilities
k8s-autopilot operates as a multi-domain deep agent system — a Supervisor routes your requests to four specialized coordinators, each managing its own team of sub-agents. Every sub-agent connects to a dedicated MCP server via Just-In-Time (JIT) connections, keeping resource usage minimal while supporting dozens of complex workflows.
Architecture at a Glance
Domain Quick Reference
| Domain | Coordinator | Sub-Agents | MCP Servers | What it Does |
|---|---|---|---|---|
| 📦 Helm Operator | helm-operator-coordinator | 7 agents | Helm MCP, GitHub MCP | Chart generation pipeline, live Helm operations, GitHub persistence |
| 🔄 App Operator | app-operator-coordinator | 3 agents | ArgoCD, Argo Rollouts, Traefik MCP | GitOps, progressive delivery, edge routing |
| ☸️ K8s Operator | k8s-operator-coordinator | 1 agent | Kubernetes MCP | Cluster ops, pod debugging, scaling, RBAC inspection |
| 🔭 Observability | observability-coordinator | 5 agents | Prometheus, Alertmanager, OTel, Loki, Tempo MCP | Full telemetry monitoring, alerting, pipelines, logs & distributed tracing |
Sub-Agent Reference
Every sub-agent follows a consistent pattern: Classify → Discover → Plan → Execute → Verify. State-modifying operations are always gated by Human-in-the-Loop approval.
| Domain | Sub-Agent | MCP Server | Connection | HITL Gated |
|---|---|---|---|---|
| Helm | helm-planner | — | Compiled Subgraph | No |
| Helm | helm-skill-builder | — | Static Dict | No |
| Helm | helm-generator | — | Static Dict | No |
| Helm | helm-validator | — | Static Dict | No |
| Helm | helm-updater | — | Static Dict | No |
| Helm | helm-operation | helm_mcp_server | JIT MCP | ✅ |
| Helm | github-agent | github_mcp | JIT MCP | No |
| App | argocd-onboarder | argocd_mcp_server | JIT MCP | ✅ |
| App | argo-rollouts-onboarder | argo_rollout_mcp_server | JIT MCP | ✅ |
| App | traefik-edge-router | traefik_mcp_server | JIT MCP | ✅ |
| K8s | k8s-cluster-ops | kubernetes_mcp_server | JIT MCP | ✅ |
| Observability | prometheus-operator | prometheus-mcp-server | JIT MCP | ✅ |
| Observability | alertmanager-operator | alertmanager-mcp-server | JIT MCP | ✅ |
| Observability | opentelemetry-operator | opentelemetry-mcp-server | JIT MCP | ✅ |
| Observability | loki-operator | loki-mcp-server | JIT MCP | No (Read-Only) |
| Observability | tempo-operator | tempo-mcp-server | JIT MCP | ✅ (CRDs only) |
Shared Infrastructure
All sub-agents share a common foundation:
- State Persistence: PostgreSQL checkpointer persists agent state across long-running workflows and HITL pauses
- A2UI Buffer Middleware: Intercepts large telemetry payloads and streams interactive visual components without prompt context degradation
- Context Engineering: Layered middleware (
SupervisorContextMiddleware,SummarizationMiddleware,ObsOperationContextMiddleware) preserves cross-domain context - Operations Journaling: Every state-modifying action is journaled in
operations-log.md