delegate-claude-adversarial-review

name	delegate-claude-adversarial-review
description	Delegate a steerable adversarial review of current changes from a Codex session to Claude Code. Hands an adversarial-review to Claude via `claude -p` and returns the challenge to Codex.

Delegate Adversarial Review to Claude

Hand off a steerable adversarial review to Claude Code from a Codex session — pressure-tests design choices, assumptions, and alternative approaches.

When to use

Use this skill when the user explicitly wants an adversarial / challenge review done by Claude (not Codex itself).

Expected prompt shape:

• $delegate-claude-adversarial-review challenge the design of the current changes
• $delegate-claude-adversarial-review have Claude pressure-test this branch

The user may include focus text after the trigger.

Instructions

1. Capture any focus text from the user's request.

2. Run Claude Code non-interactively. Adversarial review challenges direction, not just code details.

   claude -p 'Run an adversarial review of the current uncommitted changes and the current branch vs main. Read-only. Be skeptical and steerable: pressure-test design choices, hidden assumptions, tradeoffs, alternative approaches, failure modes (auth, data loss, races, rollback, reliability). 1) Run `git status`, `git diff`, `git log main..HEAD`. 2) Read AGENTS.md and related ADRs. 3) Ask: was this the right approach? Would a different design be safer or simpler? What edge cases are missed? What hidden coupling exists? 4) Print findings to stdout in a structured format (Direction Critique / Hidden Assumptions / Failure Modes / Alternatives Worth Considering). Do NOT modify any files. Focus area (optional): <focus>'
   

3. Capture stdout and summarize the adversarial findings.

4. This is a challenge — the user decides whether to accept, defer, or push back.

5. Do NOT auto-apply suggestions.