All posts
Muhtalip Dede profile photoMuhtalip Dede · Founder of kprompt

Break a kind cluster on purpose, then watch the Observe agent

One command from kprompt-examples: kind up, seven failure scenarios, verify they actually broke, then run the Observe agent offline in heuristic mode.

v0.5 shipped the optional Observe agent. The missing piece for a live walkthrough was a payments namespace that actually misbehaves — not a slide claiming CrashLoopBackOff. kprompt-examples is that fixture set.

One command

up → break-all → verify → agent-full (~45s)

git clone https://github.com/kprompt/kprompt-examples.git
cd kprompt-examples
make walkthrough

Needs Docker, kind, kubectl, and kprompt v0.5+ (the agent subcommand). Heuristic mode — no LLM API key, no spend. DEMO_SECONDS=60 stretches the agent window for recordings.

What you should see

  • CrashLoop, ImagePull, OOM, stalled rollout, unbound PVC, failing CronJob, missing Redis hostname
  • Health score move while baseline web stays Ready (so the score has something healthy to weigh against)
  • Correlated incidents and gated alerts — not one Slack message per kubelet Event
  • With --autopilot-propose: a propose-only rollback suggestion on the stalled checkout rollout (Applied stays false)

Honest limits

Heuristic analysis is deterministic and offline — useful for demos and CI, not a substitute for a real LLM run. Autopilot remains propose-only. Redis/Postgres in the dependency scenario are busybox stubs for Service-name discovery, not databases. Full caveats live in the examples README.

Related posts