ng/Adversarial Code Review

Multi-model adversarial code review — Optimizer and Skeptic agents challenge each other, auto-fix consensus issues.

Trust Signals

workflow.ymlSHA-pinned

uses: ng/adversarial-review@89624fa031bd3740ba5adb6eeec7497fc31ff06d # v1.6.1

tags can be moved; commit SHAs can't. why a SHA?

name	description	required	default
claude_code_oauth_token	Claude Code OAuth token (from Anthropic Console)	yes	—
pr_number	Pull request number to review. Defaults to the triggering PR.	no	${{ github.event.pull_request.number }}
mode	Review mode: "no-fix" (default) reports only, "auto-fix" applies consensus fixes. Any unrecognized value falls back to "no-fix" (fail-safe).	no	no-fix
paths	Comma-separated paths/globs to scope the review to (passed as --paths). Empty reviews the entire branch diff.	no	""
allowed_tools	Additional allowed tools (comma-separated). git, gh, and common build tools are included by default.	no	""
model	Model to use for the lead agent (e.g. "opus", "sonnet"). Defaults to the claude-code-action default.	no	""

no outputs