What problem does it solve? Teams running multi-agent workflows often add agents or parallelize prematurely, wasting context tokens and creating race conditions without improving throughput. This Skill forces an evidence-based review of review quality, bottleneck location, and work independence before any scaling change is recommended. ## Core Features & Use Cases - Three-question diagnostic: Checks defect detection rates from .crew/metrics.md, identifies where work actually waits, and verifies whether work units can genuinely run in parallel. - Tiered role ladder: Maps roles (explorer, security, developer, dba, planner, and more) to tiers 0-3 with explicit trigger conditions for each addition, kept in sync with crew_state.ROLE_TIERS via a committed test. - Cost accounting: States the per-role context cost on every invocation so growth decisions are weighed against measurable value in metrics. - Use Case: A user asks to "add three more developer agents" to speed up a single repo. The Skill checks the metrics, finds the bottleneck is human review attention, and recommends against the change instead of blindly scaling. ## Quick Start Ask the agent to review whether the current crew setup should grow, shrink, or parallelize based on recent metrics.
