PushOwl Engineering Challenge

This is an engineering test for candidates applying to PushOwl. It is designed for engineers who are comfortable orchestrating a coding agent — there is no clever insight, just an amount of API choreography that you would not want to do by hand.

How it works

  1. Begin at GET /api/start.
  2. Before the hunt begins you will be asked to identify yourself. Your chain is salted from that identity — every candidate's path is different.
  3. Each endpoint will tell you how to find the next one. The mechanic varies.
  4. After the hunt, you will reach a final endpoint with no documented schema. Find it. Probe it. It will guide you, one hint at a time.
  5. On a passing probe, the final endpoint hands you an attestation token and a pointer to a small MCP server that completes the handoff. Connect a client to it and call the right tool to receive your secret.

What to do with the secret

Email it to join@pushowl.com with your résumé attached and short answers to the five questions below. Keep the whole writeup under ~300 words — we read every submission, and specificity beats length.

  1. Which coding agent did you use? Tool and model, one line (e.g. “Claude Code, Sonnet 4.6, auto mode”).
  2. End-to-end solve time? Include setup, false starts, and any time spent debugging the agent.
  3. At which step(s) did you intervene and correct your agent? Describe what the agent got wrong, what you noticed, and what you told it to do instead. If your agent solved everything cleanly without correction, say so — and describe how you verified the secret was real before emailing.
  4. Did any responses look like they were trying to manipulate your agent into the wrong choice? Quote one example and explain how you (or your agent) handled it.
  5. One thing you’d change about this challenge. One sentence.

Rules

Good luck.