I have been looking at how AI Skills / agents are usually presented. A lot of pages say what the tool can do, but not enough of them show the operational case: what happens when the input is vague, the claim has weak evidence, the handoff is messy, or the release package is risky.
Tangke Creative Research has published a set of 16 Skills on SkillHub. The interesting part is that each Skill is described through a concrete service case, not only a capability list.
A few examples:
- Bianzhen checks evidence. For a claim like "this AI tool improves conversion by 300%", it asks about the product, conversion definition, baseline, sample size, time window, incentive, and third-party A/B evidence.
- Sunmao turns vague acceptance criteria into measurable ones: CTR, first screen load time, interaction latency, error rate, test evidence, and delivery proof.
- Tongjing audits a Skill / ZIP package before release: structure, static safety, declaration consistency, docs, packaging hygiene, and possible leaked credentials.
- Chuandeng builds a handoff package: goal, completed work, files, commands, test evidence, blockers, and the next first action.
- Shapan simulates the first-time user journey before launch: what a new user understands in 5 seconds, 15 seconds, and 30 seconds.
The full set also covers opportunity discovery, public feedback monitoring, naming checks, copy review, boundary protection, multi-perspective review, relationship dynamics, and deeper business judgment.
For agent builders, I think the useful idea is this: a Skill becomes much more valuable when it carries the judgment, evidence, acceptance criteria, risk boundary, and handoff format around the generation itself.
SkillHub page:
https://skillhub.cn/enterprise/org-j3zmzop1?publisher=%E5%94%90%E5%8F%AF%E5%88%9B%E7%A0%94
Top comments (0)