What problem does it solve? Complex tasks like multi-lens code reviews, audits, and migrations often need several skills applied in parallel, but running them in one context blurs each lens and produces unverified, prose-only results. This Skill turns a library of installed skills into a coordinated agent graph where each sub-agent node loads exactly one skill, works a bounded scope, and returns schema-shaped findings that are independently verified before synthesis. ## Core Features & Use Cases - Skills-as-nodes orchestration: Design a graph where each node is a sub-agent loading one skill, with typed edges (needs, informs, excludes) and stage patterns like fan-out, pipeline, adversarial verify, and judge panel. - Structured contracts and verification: Every stage returns schema-shaped data with file:line evidence and fixed severity levels; independent verifier nodes attempt to refute each finding, and unverifiable claims are labelled rather than dropped. - Three runtimes: Run via the Workflow tool (dynamic workflows with schema enforcement and resume), Agent-tool fan-out, or a sequential degraded mode when no sub-agent mechanism exists. - Use Case: Reviewing a large pull request through five architectural lenses — scout the diff inline, fan out one node per lens, dedup findings, verify each adversarially, and deliver one ranked, evidence-linked report. ## Quick Start Ask the agent to use graph engineering to fan out one sub-agent per relevant skill over your target scope, verify the findings adversarially, and synthesize a single ranked deliverable.
