Your Agent Needs a Stop Condition: A Step‑by‑Step Build

说明与用途

构建了一个停止门脚本，通过检查限制和置信度来防止代理执行不必要的操作。

问题

Agents can keep working after a task is finished, potentially making unwanted changes to files.

结果

The stop gate script successfully intercepts unwanted agent actions, providing a clear human‑review step and preventing accidental file modifications.

工作流

• 1. Clone MonkeyCode repo and obtain free access. 2. Create a project directory and install OpenAI SDK. 3. Write gate.mjs with system prompt, limits, and checks. 4. Run the gate script with a task argument. 5. If the agent outputs STOP or violates limits, hand back to human; otherwise, output is safe to show.
• 6. Verify the gate by testing with various tasks and observing the console logs.

证据

An agent summarized it, then rewrote it. Nobody asked for a rewrite.

来源

The fix is not a better prompt. The fix is a boundary you can test.

来源