Параллельные агенты для исследования кодовой базы

★ 8.5 · vibe-coding

sciomc is a Claude Code skill that orchestrates parallel scientist agents for comprehensive multi-stage research workflows, with an optional AUTO mode for fully autonomous execution. Given a research goal, it decomposes the work into 3–7 independent stages and fires each as a separate `scientist` subagent with smart model routing: haiku for data gathering and file enumeration, sonnet for code pattern analysis, and opus for complex architectural reasoning. Once parallel execution completes, a cross-validation stage checks findings for contradictions and gaps before synthesizing everything into a final report. Session management commands — `status`, `resume`, `list`, and `report` — let you track and continue interrupted research. It suits deep codebase analysis, security audits, error-handling reviews, and any investigation where parallel coverage of multiple hypotheses is needed.