A developer shares practice of using an AI agent to automatically test and submit bug reports

A developer needs to test Kerykes, a project that generates cover art and promo reels for a blog. There are no external users to catch bugs, so the developer describes using another agent session to drive Kerykes from the command line, throw inputs at it, and write up failures as issues. In a post published on DEV Community on November 7, 2026 at 10:52, the author recounts this workflow. Kerykes had collected seven bug reports over the first two days of August, and every one of the seven was filed by a machine rather than by a human: specifically, by another agent session that drove Kerykes from the shell, hit things that did not work, and then documented them. Kerykes has no external users unless you count the author themself. The developer had used a session of Keryx, a tool for creating cover art and promotional clips, to act as a kind of self-testing harness, launching it via the command line and letting it interact with its own outputs. The agent surfaced issues along the way and turned them into tickets, closing a loop in which a bot both generates content and polices its correctness. The seven tickets, however, landed in three rather different buckets. They also arrived in three fairly different pileso (原文片段在此中断). This is the only verified report found so far, and there are no later updates or confirmations to report.