claude-adversarial-review

name	claude-adversarial-review
description	Compatibility alias for stricter Claude review; prefer claude-review with --adversarial for new use.

Claude Adversarial Review

Use claude-review with --adversarial for new stricter reviews. This compatibility skill remains available for parity with earlier plugin versions.

Use the companion script from the installed plugin root. If the current working directory is not this plugin checkout, resolve the script relative to this skill file, for example ../../scripts/claude-companion.mjs, and pass --cwd <target-repo> for the repository being reviewed.

Real adversarial review calls send diffs and review prompts to Claude Code and may spend quota. If the Codex host offers persistent approvals, prefer approving the narrow prefix node scripts/claude-companion.mjs review and running review --adversarial instead of approving a separate adversarial command. If host policy blocks external disclosure, do not bypass it.

Review uncommitted changes:

node scripts/claude-companion.mjs review --adversarial

Review a branch diff:

node scripts/claude-companion.mjs review --adversarial --base main

Legacy compatibility alias:

node scripts/claude-companion.mjs adversarial-review --base main

For machine-readable output:

node scripts/claude-companion.mjs review --adversarial --base main --json

This uses the same schema, diff collection, and JSON parsing as claude-review, but with a stricter prompt that challenges assumptions and looks for concrete failure modes. It always uses Claude Code --permission-mode plan and should not edit files.

It also uses the same single-review diff limit as claude-review: 200000 bytes by default, configurable with --max-diff-bytes <n>.

Use --effort <level> to pass Claude Code effort: low, medium, high, xhigh, or max.

Model guidance:

• Use --model opus for serious adversarial review, merge readiness, or high-risk design validation.
• Use --effort low only for smoke tests, cheap sanity checks, or explicitly low-effort requests.
• Do not assume short Claude model names always map to the expected backend model. After real Claude calls, inspect and report actual model usage from JSON output when available.