HR Skills: Good Content, Bad Routing
The HR batch is a clean demonstration that content quality and selection quality are different problems.
| Skill | Trigger | Structure | Workflow | Content | Engineering | Security | Total |
|---|---|---|---|---|---|---|---|
| job-description | 7.5 | 8.5 | 8.0 | 9.0 | 8.5 | 9.5 | 8.4 |
| onboarding-plan | 7.5 | 8.5 | 8.0 | 9.0 | 8.0 | 9.5 | 8.4 |
| applicant-screening | 7.0 | 8.0 | 5.5 | 8.5 | 8.0 | 9.5 | 7.5 |
| cover-letter | 1.5 | 8.0 | 6.5 | 8.5 | 5.0 | 9.5 | 6.4 |
The cover letter anomaly
cover-letter has a content score of 8.5. The guidance is specific, the structure is sound, and the output is defined. But the trigger score is 1.5. The file does not clearly say when it should be selected.
In a directory of 800+ skills, that is a fatal routing problem. An agent scanning descriptions has no reliable reason to pick this skill over a generic writing skill. The content never gets a chance to matter.
Screening's workflow gap
applicant-screening scores 7.5 overall but only 5.5 on workflow. The content is detailed enough to be useful, but the procedure is not fully ordered: what to read first, how to decide, what to do when a resume is ambiguous, and how to record the decision.
What the strong ones do
job-description and onboarding-plan both score 8.4. They ask for the role context, produce a structured artifact, and include checks. Their triggers are not perfect (7.5), but they are specific enough to route.
The fix
For the cover letter, the fix is one paragraph: name the requests that should activate it, the requests that should not, and the exact phrases a user is likely to say. The content already earns an 8.5; the trigger is the only thing holding the total at 6.4.
This is the same trigger-versus-content lesson as the batch-level trigger analysis, seen at the level of a single domain.