What problem does it solve? Getting an independent, high-effort second opinion on local code changes is hard to standardize: reviewers vary in rigor, and quick self-reviews miss subtle bugs. This Skill runs a consistent, maximal-effort external review of your working tree or branch diff without ever modifying your code. ## Core Features & Use Cases - Automated scope detection: Reviews staged, unstaged, and untracked working-tree changes, or falls back to a branch diff against an auto-resolved base (origin/HEAD, main, or master). - Forced heavyweight configuration: Pins the review to Opus at xhigh reasoning effort regardless of the calling session's model, with an optional adversarial stance that tries to break confidence in the change. - Hardened read-only execution: Pipes the review brief over stdin, restricts the agent to Read/Grep/Glob tools, denies edits, shell, and network access, and fences untrusted diff content against prompt injection. - Use Case: Before merging a new feature branch, run the script with --scope branch --base develop --adversarial to get a severity-ranked findings report with a ship/no-ship verdict. ## Quick Start Ask the agent to review my current uncommitted changes with Claude for a second opinion before I merge.
