Регрессионное тестирование MCP-инструментов

★ 7.4 · testing

Оценка пользовательских историй (BAT Story Evaluation) is a Claude Code skill that compares MCP tool behavior between a target and a baseline version using pre-built and custom YAML stories with diff-based triage to catch regressions before release. Reach for it when you've just merged a fix and want to know which test scenarios are actually worth running; when you're not sure whether a change to a core module broke any user-facing behavior; when running the full story suite would take too long and you need smart selection based on what actually changed. The skill follows a strict three-step protocol: compute a git diff between the baseline tag or branch and HEAD, classify changed files (tool modules, core code, utilities), then automatically select relevant stories from a YAML catalog — or fall back to two smoke-test stories if only docs or config changed. It then designs at least one custom story targeting a specific regression hypothesis derived from the diff. Stories run first against the baseline, then the target version, each verified immediately via ha_query.py, with results written to a JSONL file.