Instructions files people actually ship with: rendered, split into modules you can copy one at a time, and tracked across revisions.
Fifteen lines, ~180 tokens, and it outperforms the 400-line files it replaced. Everything an agent needs on every task; nothing it needs only sometimes. The controversial part is what's missing.
Server components by default, one blessed data-fetching pattern, and hard rules about 'use client'. This file exists because every one of those rules was violated by an agent first.
The root file that keeps five agents from treating 40 packages as one big sandbox: affected-only commands, module boundaries with an owner column, and a hard rule about generated code.
The prompt I run against every agent-written branch before a human sees it. Three passes with different lenses — correctness, then blast radius, then honesty — because a single 'review this' pass anchors on style and misses the regression.
The AGENTS.md running my shipping iOS app. Simulator names, the exact xcodebuild invocations, SwiftUI conventions the agent kept violating until they were written down, and a definition of done that includes a green test run.
A scoped .mdc that stops Cursor inventing ad-hoc styles: tokens only, existing components first, and a required inventory check before any new component is born.
Agent instructions written by someone whose job is extracting secrets from agents. Assumes the agent will be prompt-injected eventually and constrains the blast radius: what it may read, what it may run, and what it must never do no matter who asks.
Twelve years of Rails opinions distilled for an agent: fat models within reason, no service-object cargo cult, bin/rails test as the heartbeat, and migrations treated like the loaded weapons they are.