Adversarial

Use this skill to challenge a plan, claim, implementation, or review from the strongest useful opposing position. This is not a default security scan and not a generic model vote. The main agent (orchestrator) remains final owner of decisions, edits, synthesis, and verification.

This skill narrows the original multi-agent security analyzer into opposite-analysis. It is harness-agnostic: the main agent — whether Claude Code or Codex — stays the decision owner. It defaults to assumption testing, counterargument generation, failure-mode discovery, and concise next actions.

Trigger

Use when the user says adversarial, analisis opuesto, opposing analysis, contrarian review, red-team this, challenge this plan, strongest counterargument, security review of this change, or asks to consult Z.ai and MiniMax.

Core Workflow

1. State the claim or plan being challenged in one sentence.
2. List the assumptions it depends on.
3. Ask what would make each assumption false in the current repo/runtime.
4. Identify the highest-impact failure modes first.
5. Separate proven issues from plausible risks.
6. Recommend one concrete next action: keep, adjust, defer, validate, or revert.