review: an agent checks the agent

fresh eyes, on demand

Verification, the module 1 habit, asks: does the change do what the request said? Review asks a harder question: what did the change break that nobody thought to check? The session that wrote the code answers the first question well and the second one badly, for the same reason you proofread your own writing badly. It trusts its approach.

A reviewer agent starts without that trust. The /code-review command reads your diff in the current session and "reports correctness bugs and reuse, simplification, and efficiency cleanups". You run it before pushing, on the working diff, and fix what it finds while the context is warm.