🧰 DevOps Interview Quest

A gamified study companion for DevOps / SRE interview preparation — Kubernetes, OpenShift, and SRE practice. Built on the quest-engine extracted from MCP Architecture Quest: concept cards with three-depth deep dives, domain-mapped quiz levels with random draws from a question bank, runbook-ordering puzzles, XP/badges, saved progress, and an AI guide in every deep dive.

No build needed to use it. index.html (once built) is a single self-contained file.

Status: scaffold

The engine is complete and working. Content is starter-level:

• 8 concept cards (Pods, Workloads, Services & networking, Storage, OpenShift deltas, RBAC, SLO/error budgets, Troubleshooting method)
• 2 runbook builders (CrashLoopBackOff, Node NotReady)
• 12 seed questions across 5 domains (core, netstore, troubleshoot, openshift, sre)

The AI guide — designed for people WITHOUT a beefy machine

Default provider is "My Claude app" (or "My ChatGPT app"): one click opens a chat in the user's own claude.ai / chatgpt.com subscription with the concept's study notes and question preloaded — no API key, no local model. The quick prompt "Interview me on this topic" turns it into a mock interviewer. Local LLM (LM Studio/Ollama) and Anthropic API remain as power options in GUIDE settings.

Develop

cd app
npm install
npm run dev          # hot-reload dev server
# ship a single file:
npx vite build && node -e '…inline dist assets…'   # see scripts/build-single.md

Growing the content (the pipeline)

1. Content lives in app/src/: data.ts (cards, runbooks, seed questions), bank.json (generated question pools), cards.json + intros.json (generated cards/intros).
2. Generate at scale with LLM agents, one per domain, each grounded in official docs (kubernetes.io/docs, docs.redhat.com OpenShift, sre.google/books). Have each agent emit JSON matching {"q","o"[4],"a",“x”,"h"?,"t"?} to /tmp/quest-bank-ops/<domain>-questions.json.
3. scripts/merge.py validates (4 options, answer index, dedupe, answer-key distribution) and writes bank.json / cards.json. Update its BANK_MAP keys to: core, netstore, troubleshoot, openshift, sre.
4. Domain emphasis on the level cards is set in app/src/questions.ts (w: fields) — tune to the target interview, not an exam grid.

Question style guidance: favor scenario stems ("a rollout is stuck…", "permission denied on OpenShift…") over definitions; add t (trap notes) for classically confused pairs (readiness vs liveness, RWO node-vs-pod, RoleBinding+ClusterRole scope).

Ideas queued

• Interviewer rubric mode: guide asks → user answers in own words → guide scores against a grounded rubric
• Terminal-output reading questions (here's kubectl describe, what's wrong?)
• More runbooks: PVC Pending, DNS failures, cert expiry, etcd alarms
• Second pack: cloud (AWS), Terraform, CI/CD

Disclaimer

Community study aid. Not affiliated with the CNCF, Red Hat, or any certification body. Verify everything against the official docs.

MIT — see LICENSE.