Using Stryker mutation testing on AI-agent code changes

On 2026-11-01 at 17:01, a tutorial titled "Agent 写的测试全绿？用 Stryker 对它改动的代码做变异测试" ("All tests green on agent-written code? Run mutation testing with Stryker on the code it changed") appeared on DEV Community under the AI category. The author is the writer of the book AI Agents Pocket Guide, the two-volume Thinking in Go series (Complete Guide to Go Programming and Hexagonal Architecture in Go), and the Hermes IDE project, an IDE aimed at developers who ship with Claude Code and other AI coding tools, with a site at xgabriel.com.

The tutorial opens with a concrete scenario: an AI agent added a discount rule and wrote two tests for it. Coverage on the agent's new file reaches 100%, every line runs, every branch runs, and both tests are green, so the author is about to merge. Then the article makes its point by deliberately breaking the source: change >= to >, and both tests still pass, because nothing checks behavior at exactly the threshold value. Swap a multiplication sign for a division sign and the suite passes as well. The takeaway is that even at 100% coverage with all tests passing, code written by an agent can still hide real bugs.

The recommended remedy is StrykerJS, used to run mutation testing scoped to the code the agent touched, deliberately corrupting the source to verify that the tests actually fail when they should. The tutorial also links out to the AI Agents Pocket Guide book, the author's other works, and the Hermes IDE project. As of the available material, this single DEV Community post is the only report on the topic, and there are no follow-up developments to record.