What problem does it solve? Reviewing an entire codebase with a single agent produces shallow, uneven coverage. This Skill fans out many focused reviewer subagents in parallel, each covering a bounded slice of the code, then synthesizes their findings into one deduplicated, severity-ranked report. ## Core Features & Use Cases - Parallel fan-out review: Splits a codebase into per-module review units (roughly 1.5k LOC each) and dispatches them as parallel subagent tasks with a shared output contract of file:line evidence plus critical/high/medium/low severity. - Batch and concurrency control: Explains the soft default of 8 parallel tasks, the absolute max of 256, and how to set concurrency so large batches run together instead of queueing. - Model selection and verification: Supports per-task model overrides and a cheap single-mode probe to confirm a model resolves before dispatching a large batch. - Synthesis and verification: Dedupes findings across reviewers, groups them by severity, writes a full REVIEW.md report, and re-reads cited lines to verify surprising Critical claims. - Use Case: When asked to audit a whole repository, launch 12 reviewers (6 for the Rust core, 3 for the Go TUI, plus SDK, web, and build/CI) in one parallel call, then deliver an executive summary with a prioritized fix list. ## Quick Start Review this entire codebase by fanning out parallel reviewer subagents per module and synthesize the findings into a severity-ranked report.
