Coding agent
Give workbench agents the ability to read and write code in your repositories
Overview
The coding agent extends a workbench with the ability to read source code, propose changes, and open pull requests. When enabled, the agent can go from "I found a problem" to "here is a PR that fixes it" in a single job run — without manual intervention.
Coding capabilities are optional. Workbenches that are purely operational (querying metrics, triaging alerts, summarizing Kubernetes state) do not need a coding agent configured.
Prerequisites
The coding agent runs inside a sandboxed Kubernetes pod managed by an AgentRuntime resource on your management cluster. You must deploy and configure an AgentRuntime before enabling coding in a workbench.
See Configure an AgentRuntime for full setup instructions, including:
- Deploying the
AgentRuntimeCRD with your AI provider credentials (Claude, OpenCode, or Gemini) - Tuning pod resource requests and limits
- Configuring network policies for the agent namespace
Once an AgentRuntime is deployed and set as default: true (or you have at least one runtime available), it will appear in the workbench coding agent step.