Выбор стратегии оркестрации агентов

orchestration

vc-agent-strategy-compare is a Claude Code skill that evaluates four execution strategies — sequential, parallel-subagents, workflow, and agent-team — for any RIPER-5 phase or fan-out task. It computes a 7-signal score table, shows explicit agent-count math, applies cost guards, and outputs a ranked strategy recommendation. Two modes are available: Simple Mode scores signals from existing conversation context, while Deep Mode runs targeted bash scans (find, grep) across the codebase before scoring each signal with concrete evidence. Model selection follows a strict policy: all spawned agents default to sonnet, with opus reserved exclusively for agents performing real code writing or build execution. The skill is invoked at five RIPER-5 checkpoints — from RESEARCH through UPDATE PROCESS — to determine the optimal orchestration approach for the next phase.