What problem does it solve? Choosing how to orchestrate multiple AI agents for a phase or fan-out task is error-prone: teams default to parallel subagents even when a sequential run, deterministic workflow, or coordinating agent team fits better, wasting tokens or losing coordination. This Skill scores the task against 7 signals and outputs a ranked strategy recommendation with explicit agent-count math and cost guards. ## Core Features & Use Cases - 7-Signal Scoring: Scores multi-package scope, schema/API surface, viable directions, phase-program classification, user depth requests, high-risk classes, and blast-radius file count to produce a 0-7 score mapped to a strategy threshold. - Four Strategy Comparison: Always evaluates sequential, parallel subagents, dynamic workflow, and agent team side-by-side with agent count math, cost guards (>30 and >100 agents), and best-fit criteria. - Simple and Deep Modes: Scores from existing conversation context by default, or runs targeted codebase scans (touchpoint files, test infrastructure, phase dependencies) before scoring when signals are ambiguous. - Use Case: Before spawning agents to validate 5 phase plans in a RIPER-5 workflow, run this Skill to get a scored recommendation showing whether parallel subagents or an agent team is correct, plus the expected agent count. ## Quick Start Ask the assistant to run vc-agent-strategy-compare on your current phase or fan-out task, for example: compare execution strategies for validating these five phase plans.
