Rails testing on autopilot: Building an agent that writes what developers won't

Mistral’s open‑source coding assistant Vibe was extended to create an autonomous agent that writes or improves RSpec tests for Ruby on Rails applications. The agent parses five main file types—models, serializers, controllers, mailers, and helpers—each of which follows a distinct testing convention. By mapping source files to spec files in a nearly 1:1 relationship, the agent can locate or identify missing tests. To avoid duplicating code, the agent manages shared resources such as factories and fixtures, creating new factory files when absent and reusing existing ones with caution. The system relies on a repository‑level AGENTS.md file that supplies a step‑by‑step execution plan and a forced self‑review loop, ensuring every public method is covered. Separate skill files for each file category, such as a dedicated controller skill, provide precise instructions that improve test quality. Vibe’s built‑in tools for file manipulation and command execution are extended with custom tools: a RuboCop linter that detects style violations and a SimpleCov tool that runs the generated spec, reports pass/fail results, and measures coverage. The agent iteratively corrects failures, achieving a high pass rate after a few iterations. The approach demonstrates how AI can automate repetitive testing tasks, reduce debugging effort, and maintain high code quality in large Rails monoliths.