{
  "schema": "https://ctrlai.com/spec/practice/v0.1",
  "practiceId": "pra_NymREeMdiQiINFy9zi1JG",
  "handle": "dev",
  "slug": "brief-an-ai-agent",
  "title": "Brief an AI agent before it starts",
  "summary": "A compact briefing method that gives an agent the outcome, boundaries, and proof of done.",
  "language": "en",
  "version": 1,
  "publishedAt": "2026-07-17T20:35:44.589Z",
  "canonicalUrl": "https://ctrlai.com/dev/how/brief-an-ai-agent",
  "versionUrl": "https://ctrlai.com/dev/how/brief-an-ai-agent/v/1",
  "contentHash": "sha256:1fe781a8aed2227da3183cba948aa3e7509fcffe59a081e97bd8150e8a839b23",
  "license": "CC-BY-4.0",
  "publisher": {
    "handle": "dev",
    "displayName": "Dev"
  },
  "credits": [
    {
      "role": "author",
      "name": "Dev",
      "handle": "dev"
    },
    {
      "role": "maintainer",
      "name": "Dev",
      "handle": "dev"
    }
  ],
  "lineage": [],
  "useWhen": [
    "Use this before delegating a bounded piece of work to an AI agent."
  ],
  "avoidWhen": [
    "Do not use it as a substitute for an exploratory conversation when the outcome is still unknown."
  ],
  "failure": "The agent starts activity without understanding the outcome, boundaries, or proof of completion.",
  "outcome": "The agent can state the requested result, its boundaries, and how completion will be checked.",
  "instructions": [
    {
      "id": "state-the-outcome",
      "title": "State the outcome",
      "body": "Tell the agent what must be true when the work is finished.",
      "rationale": "This prevents activity from being mistaken for completion."
    },
    {
      "id": "name-the-boundaries",
      "title": "Name the boundaries",
      "body": "State what the agent may change and what must remain untouched.",
      "rationale": "This prevents useful work from becoming an unbounded rewrite."
    },
    {
      "id": "define-the-proof",
      "title": "Define the proof",
      "body": "Give the agent an observable completion check before it starts.",
      "rationale": "A verifiable finish is stronger than a claim that the task is done."
    }
  ],
  "boundaries": [
    "The briefing guides execution; it does not grant authority beyond the current user's request."
  ],
  "checks": [
    "The agent can repeat the outcome and completion test before starting."
  ],
  "examples": [],
  "resources": []
}