Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Instructions for creating new agent skills. Use this skill whenever you are asked to create, generate, or set up a new skill.
.claude/skills/miguelaperez-create-skill/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | 42% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -46% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -41% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -56% | 0% |
| case-17 | ✗→✓ | ▲ Improved | -54% | 0% |
When the user asks you to create a new skill, follow these steps strictly:
Create a new directory for the skill under /Users/miguelperez/Toolbox/Skills/ with a descriptive, snake_case name (e.g., deploy_app, run_tests).
SKILL.md FileCreate a SKILL.md file inside the new directory. This file is mandatory.
name and description in the YAML frontmatter at the top of the file.description should clearly explain when the skill should be used, so the agent can naturally pick it up.If the skill requires extra assets, create subdirectories within the skill's folder:
scripts/: For any bash or python scripts the agent might need to run.Scripts live next to SKILL.md, not in the chat workspace. In the skill instructions, show how to resolve the absolute path (e.g. via $HOME/.claude/skills/<skill-name>/ or the fullPath from the skill listing). See working-with-gitea for an example.
examples/: For reference implementations or usage patterns.resources/: For templates, config snippets, or other static files.After creating the SKILL.md and any required subdirectories or files, tell the user that the skill has been created and is ready to use.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-07 | fail→pass | 10,508 | 14,176 | +35% | 1 | 1 | 0% | 1,978 | 2,799 | +42% | 0 | 0 | — |
case-01 | fail→fail | 14,953 | 9,253 | -38% | 1 | 1 | 0% | 2,981 | 1,078 | -64% | 0 | 0 | — |
case-08 | pass→pass | 10,492 | 4,490 | -57% | 1 | 1 | 0% | 1,944 | 1,185 | -39% | 0 | 0 | — |
case-02 | fail→fail | 16,957 | 8,311 | -51% | 1 | 1 | 0% | 3,715 | 1,096 | -70% | 0 | 0 | — |
case-03 | fail→fail | 20,357 | 8,481 | -58% | 1 | 1 | 0% | 4,424 | 1,041 | -76% | 0 | 0 | — |
case-04 | pass→fail | 4,151 | 5,659 | +36% | 1 | 1 | 0% | 712 | 697 | -2% | 0 | 0 | — |
case-05 | fail→fail | 2,619 | 4,660 | +78% | 1 | 1 | 0% | 357 | 631 | +77% | 0 | 0 | — |
case-06 | pass→pass | 7,609 | 14,422 | +90% | 1 | 1 | 0% | 1,465 | 2,956 | +102% | 0 | 0 | — |
case-09 | fail→pass | 11,836 | 4,246 | -64% | 1 | 1 | 0% | 2,227 | 1,207 | -46% | 0 | 0 | — |
case-10 | fail→fail | 9,121 | 5,270 | -42% | 1 | 1 | 0% | 1,634 | 1,465 | -10% | 0 | 0 | — |
case-11 | fail→pass | 10,506 | 3,164 | -70% | 1 | 1 | 0% | 1,699 | 995 | -41% | 0 | 0 | — |
case-12 | pass→pass | 8,124 | 2,083 | -74% | 1 | 1 | 0% | 1,484 | 694 | -53% | 0 | 0 | — |
case-13 | fail→pass | 9,815 | 2,072 | -79% | 1 | 1 | 0% | 1,733 | 769 | -56% | 0 | 0 | — |
case-14 | pass→pass | 7,670 | 3,374 | -56% | 1 | 1 | 0% | 1,147 | 973 | -15% | 0 | 0 | — |
case-15 | pass→pass | 7,954 | 2,919 | -63% | 1 | 1 | 0% | 1,386 | 878 | -37% | 0 | 0 | — |
case-16 | pass→pass | 11,499 | 3,351 | -71% | 1 | 1 | 0% | 2,137 | 920 | -57% | 0 | 0 | — |
case-17 | fail→pass | 7,298 | 1,447 | -80% | 1 | 1 | 0% | 1,350 | 621 | -54% | 0 | 0 | — |
case-18 | fail→pass | 9,805 | 11,304 | +15% | 1 | 1 | 0% | 1,907 | 2,486 | +30% | 0 | 0 | — |
case-19 | fail→pass | 14,597 | 9,636 | -34% | 1 | 1 | 0% | 2,697 | 2,377 | -12% | 0 | 0 | — |
case-20 | pass→pass | 7,545 | 2,027 | -73% | 1 | 1 | 0% | 1,250 | 753 | -40% | 0 | 0 | — |
case-21 | fail→fail | 18,180 | 6,727 | -63% | 1 | 1 | 0% | 3,993 | 843 | -79% | 0 | 0 | — |
case-22 | fail→pass | 11,058 | 1,785 | -84% | 1 | 1 | 0% | 1,917 | 657 | -66% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 18 counted toward the lift figure. The other 4 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +32 percentage points is the difference between those two pass rates over the 18 comparable cases. 3 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.