From the HackerLinks archive

Zork-Bench

Interactive-fiction benchmark for testing agent runtime, harness, and prompting choices.

At a glance:

First seen:2026-07-13

Last seen:2026-07-13

Times seen:1

Website:github.com

The short version

Interactive-fiction benchmark for testing agent runtime, harness, and prompting choices.

Why it caught our attention

It offers a concrete, inspectable environment for experimenting with agent behavior.

Where it surfaced on Hacker News

2026-07-13

Editorial paraphrase

A builder shared results and sessions from tuning a harness with Qwen3.6-27B.