adversarial-review-fast

Adversarial Review — Fast

Overview

The smallest adversarial review: one fresh subagent that has none of the main session's reasoning and assumes the change is broken. It attacks the change from two or three domains picked for this change, then checks its own findings and fixes against a checklist before it returns. It reviews only what changed since the last review when it can.

It is for a small change — up to about 500 lines, a test change, a small behavior change — and, above all, for re-checking the fixes a bigger review produced. The value is the same as the bigger panels: a reader without your context, told to prove you wrong. The saving is the panel: one agent instead of a roster plus two gates.

Three skills, one method, three sizes:

A plan is sized by what it would build. When the change under review is a plan document, count the code its tickets would touch once built, never the plan's own lines: diagrams, examples, and questions make a plan long without making the change large.

Every subagent is one-shot. The reviewer gets one self-contained task, returns its result, and ends. It never waits for a reply, a follow-up, or the user, and it is never resumed, messaged again, or continued (SendMessage in Claude Code, or the equivalent elsewhere). More work — a second look, a revised fix, a follow-up question — is a new subagent with everything it needs in its prompt, the earlier result included. A subagent left idle loses its prompt cache within minutes, so the next message re-reads its whole history at full price, and an agent that has already argued a position is no longer independent. A command that may run past three minutes (a test suite, a build) runs in the background with output to a log, and the subagent checks it with a wait capped at three minutes until it finishes: a silent wait past five minutes expires its cache just the same. Claude: do not use dynamic workflows for this. One plain subagent.