What problem does it solve? Reviewing large code changes single-handedly leads to shallow, inconsistent feedback. This Skill orchestrates multiple parallel review sub-agents to examine your own diffs from many angles, then merges and deduplicates findings into a prioritized must / imo / nits report. ## Core Features & Use Cases - Parallel bundle reviewers: Splits the diff into directory-based bundles of 5-8 files and assigns each to a dedicated sub-agent, plus a mandatory cross-file flow-integrity reviewer that catches wiring gaps, contract mismatches, and broken error propagation. - Severity triage and merging: Collects all agent findings, drops ungrounded claims, deduplicates overlapping reports, and finalizes severity using shared severity.md and output.md references from the single-pass /review-code skill. - Large-diff escalation: For diffs over 20 files or 1000 changed lines, optionally runs an additional independent reviewer focused on destructive failure scenarios. - Use Case: After finishing a feature branch touching 30 files, ask for a deep review and receive a consolidated report listing each issue with its location, perspective, and confirmed severity. ## Quick Start Ask the assistant to deeply review the current branch diff against origin/HEAD with a focus on correctness and error handling.
