xonovex-skill-data-oriented-design

Xonovex Platform Monorepo

Monorepo for Xonovex AI agent tools, workflows, and skills

AI coding agents handle prompts, tools, and code changes. What they don't manage is the environment around them: sandbox isolation, model provider routing, terminal sessions, reproducible toolchains, and orchestration at scale.

Xonovex fills this gap. It currently supports Claude Code and OpenCode as agents, with sandboxing via bubblewrap and Docker, VM-level isolation via gVisor and Kata Containers, confidential computing via Confidential Containers (CoCo) with AMD SEV-SNP and Intel TDX, model routing through providers like Gemini, GLM, and GPT, workspace management with Git and Jujutsu, reproducible toolchains via Nix, and Kubernetes orchestration for running agents at scale.

The included skills are token-efficient, harness-neutral, and based on current research and best practices (Agent Skills spec, agentskills.io, agents.md).

• agent-cli-go configures sandboxes, providers, and terminal sessions, then launches the agent
• agent-operator-go orchestrates agents as Kubernetes Jobs with managed workspaces, provider secrets, shared multi-agent workspaces, namespace-level policy enforcement, network isolation, and Nix toolchain provisioning
• moon-nix-toolchain wraps every Moon task in the repository's Nix flake dev shell, giving reproducible flake-pinned toolchains in local runs, pre-commit hooks, and CI
• Skills give agents coding guidelines they follow automatically; plan-driven development with worktrees, project-instruction management, insight extraction, and skill authoring all live here as consolidated skill packages

Quick Start

Agent CLI

npm install -g @xonovex/agent-cli-go
agent-cli run --agent claude --isolation bwrap --provider gemini

The sandbox is selected by three orthogonal axes — --isolation {none,bwrap,docker} × --provision {none,nix,command} × --network {host,none,proxy} — see packages/agent/AGENTS.md.

Agent Kubernetes Operator

# Install CRDs and deploy the operator
kubectl apply -k https://github.com/xonovex/platform//packages/agent/agent-operator-go/config/crd
kubectl apply -k https://github.com/xonovex/platform//packages/agent/agent-operator-go/config/default

# Create a provider and run an agent
kubectl create secret generic gemini-credentials --from-literal=api-key='your-key'
kubectl apply -f - <<EOF
apiVersion: agent.xonovex.com/v1alpha1
kind: AgentProvider
metadata:
  name: gemini-provider
spec:
  displayName: Google Gemini
  authTokenSecretRef:
    name: gemini-credentials
    key: api-key
---
apiVersion: agent.xonovex.com/v1alpha1
kind: AgentRun
metadata:
  name: review-code
spec:
  harness:
    type: claude
  providerRef: gemini-provider
  workspace:
    type: git
    repository:
      url: https://github.com/org/repo.git
      branch: main
  prompt: "Review the codebase and suggest improvements"
EOF

Agent Plugins

Each skill is a separate plugin. Skills are applied automatically when the agent detects a relevant task — no explicit slash-command invocation needed. Skills also work cross-harness (Claude Code, OpenCode, etc.) since they follow the Agent Skills spec.

Claude Code

# Add the Xonovex plugin marketplace
claude plugin marketplace add xonovex/platform