architecture-health-scanner

# Architecture Health Scanner ## Overview This skill teaches how to interpret scanner output, not how to run the scanner. The project provides its own scanner tool (configured in `.devt/config.json` under `arch_scanner.command`). This skill turns raw findings into a prioritized remediation plan. Raw scanner output is noise. Interpretation converts noise into signal by clustering related findings, identifying root causes, and separating true problems from acceptable trade-offs. ## When NOT to Use Skip for single-file changes or bug fixes that don't touch module boundaries. If the change is isolated to one module with no cross-boundary imports, this skill adds overhead without value. ## Time Budget Quick scan: **2-3 minutes**. Full scan with triage: **5-10 minutes**. ## The Iron Law ``` NO CLASSIFICATION WITHOUT READING THE ACTUAL CODE ``` Scanner output is claims, not facts. Every finding must be verified by reading the code at the reported location before being classified as true positive or false positive. Scanner output contains false positives and context-dependent findings that cannot be classified from the report alone. A coupling warning between two modules might be an intentional design decision or a genuine violation — only reading the actual code reveals which. Misclassified findings waste developer time or, worse, dismiss real architectural drift. ## The Process ### Step 1: Verify Scanner Was Run Check that scanner output exists. The scanner command is...