Your Agent Said 'Done.' Make It Prove That Before You Believe It

说明与用途

自动化代理经常在没有证据的情况下声称完成，导致错误。实施将声明与所需证据匹配的验证阶梯可消除这些问题。

问题

Agents claim completion without providing verifiable evidence, leading to undetected failures in automated workflows.

结果

Adopting the verification ladder and checklist eliminated incidents where agents falsely reported success, ensuring that only truly completed and verified tasks are marked as done.

工作流

• Re‑read the original request and list all required items.
• Run fresh verification for each claim at the appropriate rung (e.g., build log, full test run, reproduction of bug, end‑to‑end feature test, live URL response, actual output file with timestamp).
• Attach evidence to the report and explicitly state any unverified items.
• Use the checklist to catch red flags such as missing steps, stale test runs, or soft language.
• Integrate the "verify-before-done" skill into agent workflows to enforce the ladder automatically.

证据

The fix isn’t trusting the agent more carefully. It’s refusing to accept "done" as a bare word and asking, every time, what evidence backs it.

来源

A citation-tracking job ran clean, logged success, and never created the score file it existed to produce.

来源

The verification ladder: match the claim to the evidence it actually requires, and never accept a claim pitched higher than the evidence you're holding.

来源

We packaged this exact check as the verify-before-done skill in the free Agent Starter Kit — it costs nothing to drop into an agent's workflow.

来源