Многоагентное планирование задач с консенсусом

★ 8.5 · vibe-coding

ralplan is a Claude Code skill that launches multi-agent consensus planning before any task is executed. It acts as a shorthand alias for `/oh-my-claudecode:plan --consensus`, orchestrating three agents in sequence: Planner produces an initial plan with a RALPLAN-DR summary (Principles, Decision Drivers, Viable Options), Architect reviews architectural soundness, and Critic evaluates quality — each working independently from the same fixed plan snapshot, cycling up to 5 iterations until consensus is reached. The `--deliberate` flag activates high-risk mode, adding a pre-mortem covering 3 failure scenarios and an expanded test plan (unit/integration/e2e/observability); `--interactive` inserts manual checkpoints at the draft review and final approval stages. Until the user explicitly approves execution, ralplan never edits source files, runs mutation commands, commits, or pushes — the output plan is always marked `pending approval`. Designed for teams that need structured, reviewed planning with clear tradeoff analysis before implementation begins.