The 18 Skills That Scored Below 7.0
Out of 591 fully-evaluated skills, 18 score below 7.0. Where they come from is more informative than what they scored.
| Score | Skill | Category |
|---|---|---|
| 5.5 | web-design-guidelines | Design |
| 5.6 | alphaear-signal-tracker | Finance |
| 5.7 | svg-character-animation | Developer tools |
| 6.0 | pose-library-design | Data |
| 6.3 | framer-motion | Productivity |
| 6.5 | value-prop-statements | Writing |
| 6.5 | lottie-bodymovin | Productivity |
| 6.5 | remotion-best-practices | Video |
| 6.5 | alphaear-reporter | Finance |
| 6.6 | positioning-ideas | Design |
| 6.7 | brand-guidelines | Design |
| 6.7 | prioritization-frameworks | Productivity |
| 6.7 | qiaomu-anything-to-notebooklm | Documents |
| 6.7 | alphaear-logic-visualizer | Finance |
| 6.8 | tailwind-design-system | Design |
| 6.8 | alphaear-predictor | Finance |
| 6.8 | alphaear-sentiment | Finance |
| 6.9 | theme-factory | Design |
Two categories, one family
Design and finance account for 10 of 18. Five of the finance entries belong to a single family — the alphaear skills — which means the apparent category pattern is largely one author's catalogue.
That is the same lesson as the category averages: what looks like a category signal is often an author signal.
The failure mode is recognizable
Reading the low-scoring files, they share a shape. The skill is a knowledge dump: a page of guidelines, principles, and vocabulary, with no numbered workflow the agent can execute, and no verification step. It tells the agent what good design is; it does not tell it what to do first.
Compare that with what scores well in the same categories. baoyu-diagram (9.4, design) is a set of rendering instructions with concrete output. threejs-architecture-effects (9.5, design) builds something. The knowledge is in service of an action.
The rubric scores this through workflow design, the heaviest dimension at 25 points. A skill with no workflow scores at most a fraction of those points no matter how correct its opinions are.
Why this happens in design and finance specifically
Both are domains where "knowing things" is the visible professional skill. A designer's expertise is judgment; a trader's is pattern recognition. It is natural to encode that as prose.
But an agent already has general knowledge. What it lacks is the procedural detail: which tool, in what order, with what check before declaring done. A knowledge-heavy skill adds less than an author expects, and the rubric reflects that.
What to do if your skill lands here
- Convert guidance into steps. "Prefer contrast ratios of 4.5:1" becomes "compute the contrast ratio for every text/background pair; if any is below 4.5, list it and propose a compliant alternative; re-check after changes."
- Add a verification step. Something that inspects the output before claiming success.
- Cut the philosophy. If a paragraph does not change what the agent does, it costs context and earns nothing.
The full rubric, including how workflow design is scored out of 25, is at methodology.