What problem does it solve? Whole-codebase reviews often produce vague, speculative findings or drown in false positives, making it hard to know what is actually worth fixing. This Skill performs a read-only, multi-axis review (architecture, refactoring, readability, performance, robustness, security, tests, consistency) where every finding is anchored to a concrete file:line and severity level. ## Core Features & Use Cases - Survey-first mapping: Identifies stack, entry points, and hotspots (git churn, file size, dependency concentration) before deep review, and runs non-destructive tools (type checkers, linters, audit, dead-code detectors) to ground findings. - False-positive suppression: Applies a Pre-Report Gate, Quote-the-line rule for High/Critical findings, and a blacklist of common LLM review hallucinations so only defensible issues are reported. - Structured output: Groups deduplicated findings by severity with confidence, effort, and autofix labels, plus triage groups, coverage disclosure, and an improvement roadmap. - Use Case: Ask for a full review of a legacy service before a refactor; receive a prioritized list of real defects (e.g., a swallowed fetch error at a specific line) that can be handed off to a fix loop. ## Quick Start Review this entire codebase across architecture, performance, security, and tests, and give me a severity-ranked list of findings with file and line evidence.
