What problem does it solve? Understanding an unfamiliar or legacy codebase is slow and error-prone, and ad-hoc audits often mix observations with analysis, producing findings without an honest conceptual model. This Skill performs a forensic architectural audit that builds a layered, C4-style description of what a system actually is while accumulating a separate, evidence-cited catalog of problems. ## Core Features & Use Cases - Layered architectural description: Produces C4-altitude artifacts starting with L1 System Context (prose summary, environment diagram, conceptual structure diagram, and component table), with L2–L5 layers planned for containers, components, dimensional findings, and synthesis. - Multi-agent modeling with synthesis: Spawns three independent blind modeler subagents plus a synthesizer that reconciles agreements, flags divergences, and rejects unsound proposals against hard altitude rules. - Parallel findings catalog: Maintains a running FINDINGS.md with draft IDs, file:line evidence, draft severity, and status, kept strictly separate from the pure-description artifact. - Use Case: Point the skill at an inherited repository to get a one-page system context document a newcomer can read in five minutes, plus a deduplicated catalog of issues like god objects and layering violations with concrete code citations. ## Quick Start Run /karchitect-audit with target set to the repository path and layer 1 to generate the system context description and initial findings catalog under docs/architecture/audit.
