A repo-as-agent for Kubernetes development and operations. The agent is shipped as a repository: you clone it, keep it in sync with main, and launch your coding agent from the repo root to get a Kubernetes expert that knows how to review CRDs, controllers, and more.

Clone the repository and keep it up to date with main:

git clone <repo-url> kubernetes-expert-agent
cd kubernetes-expert-agent
git pull origin main

Launch any coding agent from the root of this repository. The agent reads AGENTS.md to pick up its identity and capabilities.

Note on skill location. Skills live under .agent/skills/. Most coding agents (Codex, OpenCode, Py, and others) support discovering skills from this path. Claude Code is the exception: it reads skills from .claude/skills/, not .agent/skills/. To use this repo with Claude Code, symlink or copy .agent/skills/ to .claude/skills/, for example:

ln -s ../.agent/skills .claude/skills

The CRD / API Review scenario uses kube-api-linter (KAL) as a dependency. The agent installs it on first use; you only need to provide the build prerequisites:

• Go toolchain on PATH (KAL and the custom binary are built from Go source).
• curl (used by the golangci-lint installer).

Pinned versions live in tools/versions.env. The install script (scripts/prepare-kube-api-linter.sh) downloads the pinned golangci-lint and builds the custom KAL binary into .kube-api-linter/, which is gitignored. To upgrade KAL, bump the version there and re-run the script.

This agent is built incrementally. Current scenarios:

1. CRD / API Review - reviews a Kubernetes CRD or API (Go API types and/or CRD YAML). Runs kube-api-linter first for Go API types, then applies a manual design checklist.
2. Controller Review - reviews controller code (reconciliation, status, error handling, leader election, caching, finalizers). Accepts a local path or a Git PR link; detects intent and clarifies scope with the user.

Specific review rules for each scenario are placeholders and will be filled in progressively.