Run multi-agent code reviews and refactors for Apple platforms

Reviews a diff with four read-only sub-agents in parallel to find behavioral regressions, security or privacy risks, performance or reliability issues, and contract or test coverage gaps - purely a review skill, distinct from a companion simplify-and-fix skill: sub-agents never edit files and the main agent never applies fixes as part of this workflow. Step 1 determines scope in priority order (explicitly named files, current git changes, an explicit branch/commit/PR diff, or - only as a fallback for a review request with no clearer diff - the most recently modified tracked files), picks the smallest correct diff command for the repo state, reads local instructions (AGENTS.md, workflow docs, module architecture/contract docs) for the touched area, and builds a short intent packet for the reviewers covering what behavior is meant to change, what should stay unchanged, and any compatibility/rollout/security/migration constraints - inferring intent from the diff and flagging the inference as possibly incomplete if the user didn't state it. Step 2 launches four parallel read-only reviewers (or reviews locally for a tiny diff), each given the same scope and intent packet, forbidden from editing, patching, staging, or committing, and instructed to skip nits and speculative concerns without concrete impact. Sub-agent 1 (Intent and Regression) checks for unintended behavior drift outside the stated scope, broken edge cases or fallback paths, contract drift between callers and callees, and missing updates to adjacent flows that should change together. Sub-agent 2 (Security and Privacy) checks for missing or weakened authn/authz, unsafe input handling or injection risk, secret/token/sensitive-data exposure, and risky defaults or trust of unverified data. Sub-agent 3 (Performance and Reliability) checks for duplicate work or redundant I/O, new work added to hot paths, leaks or missing cleanup or retry storms, and ordering/race/failure-handling problems that make the change brittle. Sub-agent 4 (Contracts and Coverage) checks for API/schema/type/config/flag mismatches, migration or backward-compatibility fallout, missing or weak tests for the changed behavior, and missing logs/metrics/assertions that would make a regression harder to detect - all four reporting only issues that materially affect correctness, security, privacy, reliability, compatibility, or confidence, explicitly preferring to miss a nit over burying the user in low-value noise. Step 3 has the main agent aggregate and filter aggressively - dropping duplicates, weak/speculative claims, issues conflicting with the stated intent, and minor style comments unless they hide a real bug - normalizing survivors into file/line, category, severity (high/medium/low), why it matters, recommended follow-up, and confidence, converting ambiguous-intent findings into open questions rather than firm findings. Step 4 orders the output high-severity-high-confidence first, then medium-severity issues likely worth fixing before merge, then lower-severity follow-ups that can wait - stating plainly if nothing material was found rather than manufacturing feedback. Step 5 closes with a clear path forward grouped into fix now, fix soon, and optional follow-up, without implementing any of it.

Use it when someone asks for a review swarm, parallel review, or diff review to find behavioral regressions, security/privacy risks, performance/reliability issues, or contract/coverage gaps. It never applies fixes - for a workflow that also applies safe, high-confidence fixes after review, use the companion review-and-simplify-changes skill instead.

Input is an explicit file scope or a git diff (staged, unstaged, or an explicit comparison) plus an intent - stated or inferred from the diff. Output is a severity-ordered, evidence-backed findings report (file/line, category, severity, why it matters, recommended follow-up, confidence) and a fix-now/fix-soon/optional-follow-up recommendation - no edits applied.

Uses four parallel read-only sub-agents (reviewer role) for the four review lenses, and reads local AGENTS.md/workflow/contract docs to build the intent packet reviewers work from.

Developers and reviewers who want a fast, multi-lens (regression/security/reliability/contracts) risk assessment of a diff before merging, prioritized by severity and confidence, without the tool making any changes itself.