Amartya Gaur
01Agent systems
← The index

hunr.ai

Lets candidates use AI, then checks they understood it.

Made for
Founder · sole author
Years
Since 2026
Status
Live
Source
hunr.ai

Screening that hands candidates a real repo, lets them use any agent, then checks they can defend it. Candidate code is graded on gen2 microVMs, not gVisor, chosen because only a real kernel gives you the in-job namespace unshare that drops egress mid-run, which then needs a shim to raise loopback again or every in-process test fails on principle. Per-challenge images are built from a package list with no daemon and no build step, because an agent-authored environment should not be able to run one.

A film cut from the product’s own source

Every screen in it was lifted out of the repository, so the film cannot invent a UI that does not exist.

The numbers on it are the ones the product printed, not an illustration of them.

What it hands back

A real finished report, with leaders struck to the three decisions in it that were mine.

Fig. 1

A finished hunr report. A fit score of 87 beside a code score of 82 and an ownership score of 94; an executive summary; a line showing the fit score is the code score scaled by ownership; lists of strengths and risks; and a competency breakdown for error handling, API design and testing, each citing specific files.

  1. The arithmetic is printed. A fit of 87 is a code score of 82 scaled by 0.94, and the page shows you that line rather than asserting the 87.

  2. Every sentence names the file and line it was read from, so a hiring manager can check the claim instead of trusting it.

  3. The weakest competency stays amber on the page. Nothing is rounded up into the headline number.

What I wrote about it

  1. July 2026Proving a challenge discriminates before anyone takes itEvery hunr challenge ships with two solutions: one an expert would write, one a plausible engineer would. If the hidden tests can't tell them apart, it never goes live.
  2. August 2026A container per job, without a daemonBuilding a bespoke image for every unit of work, from a package list rather than a Dockerfile, and why having no build step at all is the security property, not a limitation.
  3. August 2026Taking the network away from a Cloud Run jobRunning untrusted code with no egress, on managed infrastructure, without a VM you have to operate. The trick is one prefix, and the reason it looks broken at first is that loopback starts down.