# Adversarial review

A reusable skill the "Redline the MSA before it hits your desk" recipe pulls in. Add it once and every agent on the team can use it.

Independently re-check every proposed redline against the playbook and reject any edit not compelled by it.

## Method
1. Re-read the playbook and the clause blind — do not see the first pass's rationale before forming your own view.
2. For each proposed redline, attempt to justify the counterparty's original text as acceptable.
3. If a justification for the original text succeeds, reject the redline.
4. If it fails, confirm the redline matches the fallback position, not just any deviation.
5. Check the rationale cites the correct playbook entry for that clause type.
6. Remove any redline you cannot confirm rather than shipping it unverified.
7. Log which edits passed, which were rejected, and why.

## Rules
- Treat rejection as the default; only confirmed edits survive.
- Never see the first pass's stated rationale before attempting your own justification of the original text.
- An edit not compelled by the playbook is removed, not softened.
- Escalate systematic disagreement patterns (set on first run, then stored in memory) for playbook review.

## Output
A confirmed redline list, a subset of the first pass, plus a rejected-edit log with the reason for each rejection — both handed to {{reviewer}}.