What 765 Taglines Reveal About How Skills Describe Themselves
We have argued that the boundary statement — what a skill refuses to do — is the strongest quality signal in a description. The corpus says almost nobody writes one.
We have 765 skills in the directory, each with a one-line tagline drawn from its source repository. Some simple corpus counts:
| Pattern | Count | Share |
|---|---|---|
| Tagline opens with "Use when…" | 175 | 22.9% |
| Tagline contains "not" or "never" | 4 | 0.5% |
| Tagline contains Chinese characters | 22 | 2.9% |
| Empty tagline | 0 | 0% |
| Duplicate taglines (exact, case-insensitive) | 3 groups covering 12 skills | 1.6% |
The first number is encouraging: nearly a quarter of authors have learned that a description is a routing instruction and open it with the trigger condition.
The second number is the finding. Four skills out of 765 tell you what they will not do.
Why the negative space is rare
Writing "not for X" requires two things most authors do not have at the moment they publish: a clear idea of the adjacent use cases their skill will be confused with, and a willingness to send that traffic elsewhere.
Both are hard. The first requires having watched the skill misfire. The second requires trusting that a narrower skill is a better skill — which is true in a router, but feels wrong when you are trying to get people to install something.
The descriptions that do include it read differently. They name specific adjacent requests and, often, the competing tool:
> Not for drafting new agreements (use an NDA generator), not for jurisdiction-specific compliance questions (use a compliance skill).
That is a skill confident enough to lose a request on purpose.
The Chinese-language subset
22 skills carry Chinese in the tagline — 2.9% of the corpus. That is a small enough group that I will not draw conclusions from it, but it is worth noting that the bilingual publishing skills (the baoyu family, 21 skills, average 9.11) are a substantial part of it. Bilingual descriptions tend to be longer, and length correlates weakly with score (r=0.24) — but the sample is too small to claim a bilingual advantage.
The duplicates
Three exact-duplicate tagline groups cover twelve skills. In each case they come from the same repository: a family of skills that share a boilerplate tagline. That is a small version of the same problem the router has — if three skills claim the same trigger, the agent has no basis for choosing between them.
What this means if you are writing a skill
The corpus gives you a cheap way to be unusual: write the sentence almost nobody writes. One line naming two adjacent scenarios your skill should not handle puts you in the top 0.5% of descriptions by a property that directly affects routing quality.
It costs a sentence. It is the highest-leverage sentence in the file.