DevOps with Claude Code

Debug the deployed app with Claude Code.

Bring deployment context into the same conversation as your code. Use Wodby task logs and runtime state to investigate a staging failure, then separate diagnosis, code changes, and deployment approval.

Connect Claude Code Follow the walkthrough

A practical walkthrough

Work toward a result you can verify.

An evidence-backed diagnosis, a reviewable proposed fix, and a clear way to verify the result.

  1. Step 1

    Identify the app and the failing task

    After connecting Wodby, run /mcp in Claude Code to confirm authentication. Name the staging app instance, expected behavior, and the task or time window to investigate.

    Check that Claude Code found the intended app instance and revision. Distinguish a build failure from a deployment, startup, or application failure.

    Claude Code step 1 prompt
    Load Wodby’s troubleshooting guidance. Identify the staging app instance I named and inspect its latest failed deployment task, steps, and logs. Compare the deployed revision with this repository. Report the first useful failure evidence. Do not edit files or change the app.

  2. Step 2

    Correlate logs with the code

    Have Claude Code compare task evidence, runtime state, and the relevant code paths. When the failure needs reproduction, ask for a brief watch of the affected container rather than an unbounded log stream.

    Look for the specific error, container, and time window. Treat logs as sensitive. Stop the watch after reproduction and distinguish confirmed evidence from hypotheses.

    Claude Code step 2 prompt
    Use Wodby’s troubleshooting guidance to investigate the failing service in my staging app. Check its latest deployment, task logs, and runtime status. If needed, watch the selected container’s logs briefly while I reproduce the issue. Do not restart, redeploy, or change the app.

  3. Step 3

    Review the fix before deploying it

    Ask for a proposed patch or configuration change and the checks that would prove it works. Review the diff and tests before separately authorizing a staging deployment.

    After an approved fix is deployed, follow its task and repeat the original reproduction. A green deployment task does not by itself prove that the incident is resolved.

    Claude Code step 3 prompt
    Based on the evidence, propose the smallest fix and explain how to validate it. Show the files or Wodby configuration that would change, tests to run, and rollback limits. Wait for my approval before editing files, changing configuration, or deploying.

Keep a short incident summary.

Record the failure, supporting task or log evidence, approved change, verification result, and remaining uncertainty. Reuse that context for the next investigation without retaining secrets or unnecessary log data.

These prompts guide the agent; they do not replace access controls. Choose scoped Wodby permissions and authorize changes deliberately.

Set up Claude CodeReview access boundaries →All agent workflows →