Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Author a high-quality Agent Skill (SKILL.md) that an AI reliably triggers and executes well — strong frontmatter, a sharp description with trigger phrases, a clear output contract, quality checks, and anti-patterns. Use when asked to write a skill, create a SKILL.md, improve a skill, review a skill for quality, or contribute to a skills library. Produces a complete, SkillCheck-passing SKILL.md plus a short rationale for the key choices.
.claude/skills/mohitagw15856-writing-great-skills/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 63% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 47% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 70% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 35% | 0% |
| case-17 | ✗→✓ | ▲ Improved | 21% | 0% |
Un skill es una promesa: dado este tipo de solicitud, produce este tipo de output profesional, siempre. Los mejores archivos SKILL.md ganan en dos cosas — el modelo los dispara en el momento correcto, y una vez disparado produce el artefacto correcto sin necesidad de intervención manual. Este skill te ayuda a escribir uno que haga ambas cosas.
Dado un concepto inicial ("un skill para escribir changelogs"), produce el SKILL.md completo de todas formas — infiere el deliverable, inputs y estructura, y marca las elecciones genuinamente abiertas. Nunca devuelvas un esqueleto con <!-- TODO --> pendiente; rellénalo.
Solicita (si no se proporcionan ya), o infiere y etiqueta:
yaml--- name: kebab-case-name # coincide con la carpeta; corto, específico description: "<una frase rica>" ---
La descripción es la línea más importante del archivo — es todo lo que el modelo ve cuando decide si cargar el skill (divulgación progresiva: solo nombres + descripciones están en contexto hasta que uno se invoca). Una descripción sólida tiene tres partes:
Escribe los disparadores como los usuarios hablan, no como los categorizarías. Cubre sinónimos.
Abre el cuerpo con una sola frase sobre el valor, en la voz de un profesional senior.
Declara que el skill entrega un artefacto completo incluso con inputs mínimos — infiere y etiqueta supuestos, nunca dejes placeholders entre corchetes, nunca rechaces por falta de contexto. Esto es lo que separa un skill que funciona de uno que demanda.
Una breve lista de qué solicitar — e instrucciones para proceder con inferencias etiquetadas si faltan.
El corazón del skill: una plantilla concreta — encabezados reales, tablas y secciones — del artefacto final. Muestra la forma, no la describas de forma abstracta. Aquí es donde vive la mayor parte de la calidad.
Una breve lista de verificación que el output debe satisfacer (la rúbrica que un revisor aplicaría). Hazlos observables.
Los modos de fallo específicos a evitar — los outputs perezosos o genéricos que un modelo más débil produciría.
npm run skillcheck (estructura) y ejecútalo contra un brief mínimo para confirmar que no pide inputs.Devuelve:
skills/<name>/SKILL.md.name está en kebab-case y coincide con la carpeta previstaTODO/placeholder dejadonpm run skillcheck sin errores<!-- TODO --> o placeholders entre corchetes en el archivo final| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 30,979 | 23,896 | -23% | 1 | 1 | 0% | 2,333 | 3,812 | +63% | 0 | 0 | — |
case-02 | fail→fail | 25,451 | 21,080 | -17% | 1 | 1 | 0% | 3,257 | 4,508 | +38% | 0 | 0 | — |
case-03 | fail→pass | 22,785 | 27,108 | +19% | 1 | 1 | 0% | 3,042 | 4,460 | +47% | 0 | 0 | — |
case-04 | pass→fail | 16,200 | 23,502 | +45% | 1 | 1 | 0% | 1,843 | 4,123 | +124% | 0 | 0 | — |
case-05 | fail→fail | 9,519 | 31,321 | +229% | 1 | 1 | 0% | 679 | 5,013 | +638% | 0 | 0 | — |
case-06 | pass→pass | 13,403 | 28,492 | +113% | 1 | 1 | 0% | 2,845 | 4,871 | +71% | 0 | 0 | — |
case-13 | pass→pass | 39,867 | 18,479 | -54% | 1 | 1 | 0% | 2,875 | 4,373 | +52% | 0 | 0 | — |
case-07 | pass→pass | 18,690 | 24,634 | +32% | 1 | 1 | 0% | 2,328 | 4,415 | +90% | 0 | 0 | — |
case-08 | pass→pass | 22,044 | 20,211 | -8% | 1 | 1 | 0% | 2,228 | 3,875 | +74% | 0 | 0 | — |
case-09 | pass→pass | 22,486 | 23,826 | +6% | 1 | 1 | 0% | 2,782 | 4,088 | +47% | 0 | 0 | — |
case-10 | fail→pass | 26,534 | 23,727 | -11% | 1 | 1 | 0% | 2,527 | 4,294 | +70% | 0 | 0 | — |
case-11 | fail→pass | 20,798 | 24,864 | +20% | 1 | 1 | 0% | 3,150 | 4,267 | +35% | 0 | 0 | — |
case-12 | pass→pass | 23,933 | 19,610 | -18% | 1 | 1 | 0% | 3,010 | 4,432 | +47% | 0 | 0 | — |
case-14 | pass→pass | 21,189 | 19,762 | -7% | 1 | 1 | 0% | 3,480 | 4,275 | +23% | 0 | 0 | — |
case-15 | pass→pass | 29,742 | 23,814 | -20% | 1 | 1 | 0% | 4,120 | 5,316 | +29% | 0 | 0 | — |
case-16 | fail→fail | 25,129 | 24,523 | -2% | 1 | 1 | 0% | 3,195 | 5,043 | +58% | 0 | 0 | — |
case-17 | fail→pass | 26,348 | 24,519 | -7% | 1 | 1 | 0% | 3,697 | 4,476 | +21% | 0 | 0 | — |
case-18 | pass→pass | 23,911 | 26,047 | +9% | 1 | 1 | 0% | 3,513 | 5,046 | +44% | 0 | 0 | — |
case-19 | fail→fail | 19,217 | 21,508 | +12% | 1 | 1 | 0% | 2,262 | 4,143 | +83% | 0 | 0 | — |
case-20 | pass→pass | 25,006 | 23,318 | -7% | 1 | 1 | 0% | 3,199 | 3,936 | +23% | 0 | 0 | — |
case-21 | pass→pass | 17,886 | 27,788 | +55% | 1 | 1 | 0% | 3,199 | 4,808 | +50% | 0 | 0 | — |
case-22 | pass→pass | 30,563 | 29,905 | -2% | 1 | 1 | 0% | 4,085 | 5,470 | +34% | 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. The headline lift of +18 percentage points is the difference between those two pass rates over the 22 comparable cases. 2 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.