claude-code-adapter

The claude-code runtime for the Language Operator: an interactive Claude Code terminal agent that runs as a native Kubernetes workload.

This repository is self-contained — it builds both the runtime image and the Helm chart that registers the claude-code LanguageAgentRuntime.

What's here

• Image (ghcr.io/language-operator/claude-code-adapter) — a combined image used by both the init container (seed-config.mjs, which translates the operator's /etc/agent/config.yaml into Claude Code's native settings) and the main container (server.mjs, the xterm.js / tmux WebSocket terminal). Ships the Claude Code CLI, the GitHub CLI, a Go toolchain, and common Unix tools.
• Chart (chart/) — renders the cluster-scoped claude-code LanguageAgentRuntime. Published to oci://ghcr.io/language-operator/charts/claude-code.

Install

Prerequisite: the language-operator chart must be installed first — it provides the LanguageAgentRuntime CRD.

helm install claude-code oci://ghcr.io/language-operator/charts/claude-code \
  --namespace language-operator

Then reference it from a LanguageAgent:

apiVersion: langop.io/v1alpha1
kind: LanguageAgent
metadata:
  name: my-agent
spec:
  runtime: claude-code

Claude Code authentication is interactive: open the agent terminal and run /login.

Development

make build      # docker build -t ghcr.io/language-operator/claude-code-adapter:latest .
make test       # build, then run the in-image smoke tests (/app/test.sh)
make publish    # build and push the image to ghcr.io

helm lint chart
helm template claude-code chart

For a local cluster, build the image and import it into k3s, then install the chart with image.pullPolicy=Never:

make build
docker save ghcr.io/language-operator/claude-code-adapter:latest | sudo k3s ctr images import -
helm install claude-code chart --namespace language-operator --set image.pullPolicy=Never

CI

• build-image.yaml — builds and pushes the image to ghcr.io on push to main and v* tags.
• release-chart.yaml — packages chart/ and pushes it to oci://ghcr.io/language-operator/charts.
• test.yaml — builds the image, runs the smoke tests, and lints/templates the chart on every PR.