Предварительная экспертиза изменений через дебаты персон

research

vc-predict is a Claude Code skill that runs a five-persona expert debate on a proposed change before any code is written, surfacing architectural, security, performance, and UX risks early. The Architect, Security, Performance, UX, and Devil's Advocate personas each analyze the proposal independently, then debate conflicts and deliver a GO / CAUTION / STOP verdict with actionable recommendations. In Deep mode a research subagent runs first — it scans git log, reads phase reports under `process/features/`, and greps CI logs for FAIL patterns, then returns a Historical Context block detailing what was tried before, what broke, and which specific lines are known landmines. The skill is designed for major features, significant refactors, and contested design decisions; it is explicitly not meant for trivial changes or already-approved work with no open design questions.