InterviewLab

AI agents

Real-world examples

Example: Debug a production bug

Prompt examples are illustrative; commands are verified against official docs.

Scenario: login returns HTTP 500 after the latest deploy. A reasonable agent-driven investigation: gather logs, form a hypothesis, reproduce locally, fix, verify.

Workflow

  1. Gather evidence (logs, recent changes, error output).
  2. Hypothesize with the agent.
  3. Reproduce the failure.
  4. Fix the root cause.
  5. Verify with tests and a re-run of the failing path.

Claude Code

Start a session and gather evidence — paste logs directly, or pipe them in non-interactively:

claude
Here is the stack trace from production: [paste]. Also run
`git log --oneline -10` and tell me which recent change could cause a 500
on POST /login.

Piped variant:

kubectl logs deploy/api --since=30m | claude -p "Find the root cause of the login 500 in these logs."

Use the built-in diagnostic skill for runtime issues:

/debug Login returns 500 after the latest deploy

Hypothesize and reproduce:

Form a hypothesis, then reproduce the failure locally by running the auth
service and posting to /login with a test user.

Fix and verify:

Fix the root cause, add a regression test that reproduces the 500, and run
the full auth test suite.

If a fix attempt goes sideways, /rewind rolls code and conversation back to a checkpoint.

Codex

Investigate interactively in the TUI, or script the log triage with codex exec (default sandbox for exec is read-only — safe for log analysis):

kubectl logs deploy/api --since=30m | codex exec "Summarize the root cause of the login 500 from these logs."

In the TUI, correlate with recent changes:

codex
POST /login returns 500 since the last deploy. Look at `git log -10` and
the auth code, and identify the most likely culprit.

Reproduce under the workspace sandbox:

Start the auth service locally, reproduce the 500 with curl, and show me
the failing code path.

Fix and verify:

Fix the root cause and add a regression test. Run `npm test -- auth` until
green, then summarize the diff.

Check the result with /diff before committing. Background dev servers show up in /ps; /stop stops them.

Kimi Code

Start a session and gather logs via shell mode (! in the empty input box runs a local command; output enters the conversation):

kimi
!kubectl logs deploy/api --since=30m | tail -100
Login returns 500 after the latest deploy. Given the logs above and
`git log --oneline -10`, what is the most likely root cause?

Reproduce:

Run the auth service locally in the background, reproduce the 500 with a
curl POST to /login, and capture the stack trace.

Fix and verify:

Fix the root cause, add a regression test that reproduces the 500, and run
the auth test suite to confirm green.

For a longer unattended investigation you can hand off a persistent objective:

/goal Reproduce the login 500 locally, fix the root cause, and finish only
when the auth test suite passes.

Last verified: 2026-09-07

AI coding tools evolve quickly. If you find outdated information, please open an issue or pull request.

View this doc on GitHub