“Make it better” is not a requirement. It is a future disagreement.
Sunmao is one of the TANCO SkillHub Skills. Its job is to turn vague delivery language into measurable acceptance criteria and evidence.
Example: “make the page better” becomes:
- CTR ≥ 8%
- first-screen load time ≤ 1.5s
- interaction feedback ≤ 300ms
- error rate ≤ 0.1%
Then it asks for the evidence package: instrumentation, Lighthouse results, stress testing, link checks, and any other verification needed for the delivery context.
This matters for AI-agent workflows because agents are good at producing artifacts, but vague acceptance criteria still produce vague outputs. A workflow needs a gate that says what “done” means before execution begins.
TANCO SkillHub:
https://skillhub.cn/enterprise/org-j3zmzop1?publisher=%E5%94%90%E5%8F%AF%E5%88%9B%E7%A0%94
Question: where do you define acceptance criteria in your agent workflow — before prompting, inside the task spec, or as a separate review step?
Top comments (0)