dont-even-try

To audit AI outputs for hidden fallbacks, you need an adversarial review process that scrutinizes execution logs and transcripts. This Skill performs a hostile inspection of prior AI actions to verify correctness, detect manipulation, and ensure accountability without altering any data.

Adversarial review in AI governance is a hostile verification process that inspects AI workflows to ensure integrity. It systematically scrutinizes execution logs and tool calls to detect hidden problems, overlooked errors, or untrusted success claims within decision processes.

Yes, you can review AI execution logs without modifying data by using a read-only inspection scope. This Skill operates exclusively with read-only commands like git diff and Get-Content to examine transcripts and diffs, ensuring the original engineering artifacts remain unaltered.

To verify the integrity of a previous AI workflow decision, conduct a systematic review of the prior turn's transcripts and tool calls. This Skill identifies untrusted success claims and hidden biases by scrutinizing the decision process for errors or manipulations.

Yes, this adversarial review approach works directly with git diff and command line logs. It uses read-only inspection tools like rg and Get-Content to analyze transcripts and execution logs, ensuring transparency and detecting hidden problems in engineering workflows.

You should use a hostile review process on AI generated code when you question the AI's output quality or honesty. It is suitable for AI governance scenarios where you need to systematically scrutinize prior turns to verify correctness and uncover potential flaws.