The Second Build Was Mostly Wiring
My first series, Your agent might actually lie to you , is about one shape of GuardianKane: a Claude Code Stop hook that will not let the agent finish until Kane CLI has checked the work in a real browser. That shape placed 2nd in the TestMuAI Kane CLI hackathon, and four paired experiments stand behind it. It is not what GuardianKane is today. The hook is still there, but it is one piece of a…
The second version of the system, referred to as V2, was primarily focused on wiring various components together. It built upon the foundation laid in Era 1, which served as a planning tool, and Era 2, which acted as a gate. Era 3 introduced several new features and improvements.
To address the issue of AI-slop tests, V2 shifted the verification process from push-based to pull-based. This meant that verification only occurred for specific nodes in the codebase, rather than generating all possible tests at once. However, the spec clarified that this approach narrowed the scope, not the depth, of the testing process.
Another significant change was the introduction of memory by the set of graph nodes under test. This allowed the system to keep track of the files involved in failures, providing more precise knowledge of which files were fixed and when.
The design of V2 also emphasized the separation of tasks between GuardianKane and the review process. GuardianKane was responsible for orchestration, including claim graph, coverage, reconciliation, and explanations. The review task, on the other hand, owned tasks, phases, the gate, and the dashboard.
One of the early fixes in V2 was the addition of a Phase 0, which implemented a review gate. This gate required every use case and generated test to be approved before they could be counted, ensuring that only rigorously verified code would be built.
Overall, V2 aimed to create a more efficient and reliable system by focusing on targeted verification, improved memory, and a clear separation of tasks between GuardianKane and the review process.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.