Guide
Writing agent policies in plain English
By Yuntong Zhang, Co-founder & Head of Research, Verigrey · PhD, National University of Singapore1 min readUpdated
A practical guide to turning fuzzy guardrails into precise, checkable sentences your whole team agrees on.
Most agent policies start as a sentence in a Slack thread — "it should never do X after Y." The gap between that sentence and something a testing platform can actually verify is where most assurance programs stall.
This guide walks through the small set of patterns (temporal ordering, data-flow constraints, tool-call preconditions) that cover the overwhelming majority of real-world agent policies, with before/after examples of vague guardrails rewritten as precise, checkable statements.