Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Write an operational runbook for a service, incident type, or deployment procedure. Use when asked to write a runbook, create an ops guide, document an operational procedure, or prepare an incident response playbook. Produces a runbook with overview, prerequisites, step-by-step procedures, rollback steps, troubleshooting table, and escalation paths.
.claude/skills/mohitagw15856-runbook-writer/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-15 | ✗→✓ | ▲ Improved | 75% | 0% |
| case-01 | ✗→✓ | ▲ Improved | 45% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -1% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 47% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 45% | 0% |
Produce runbooks operacionales para servicios, tipos de incidentes y procedimientos de despliegue — estructurados de modo que un ingeniero on-call que nunca ha tocado el sistema pueda seguirlos bajo presión.
Pide esto si no se proporciona:
Runbook: Título del Runbook] Servicio: Nombre del Servicio] Tipo: Despliegue / Respuesta a Incidentes / Mantenimiento / Recuperación ante Desastres] Última Actualización: Insertar la fecha de hoy en formato YYYY-MM-DD] Propietario: Equipo o persona] Severidad: P1 / P2 / P3 — si es tipo incidente]
Qué cubre este runbook: 1–2 frases sobre el escenario que maneja este runbook]
Cuándo usar este runbook:
high-error-rate-payment-service]main]Tiempo estimado para completar: X minutos / X–Y minutos dependiendo del resultado]
Impacto si no se completa correctamente: p. ej. Procesamiento de pagos degradado / Riesgo de pérdida de datos / Usuarios bloqueados]
Acceso requerido:
production-account]vault read secret/payment-service]Herramientas requeridas:
kubectl v1.28+]Antes de empezar:
#ops-live que estás comenzando]Enumera cada paso. Usa comandos exactos. No parafrasees nombres de herramientas o flags.
Paso 1: Nombre de la acción] Qué estás haciendo y por qué — una oración]
bash# Comando exacto [comando aquí]
Salida esperada: [qué debería aparecer si esto funcionó] Si esto falla: Mensaje de error exacto a buscar] → Qué hacer, o ver Resolución de Problemas]
Paso 2: Nombre de la acción] Misma estructura que el Paso 1]
Paso 3: Verificar Siempre incluye un paso de verificación después del procedimiento principal:
bash[comando de verificación]
Estado esperado: Cómo se ve un sistema saludable después de completar este runbook]
Cómo deshacer este procedimiento si algo salió mal:
Paso R1: Acción de reversión]
bash[comando de reversión]
Verificar reversión: [comando para confirmar que la reversión fue exitosa]
| Síntoma | Causa Probable | Resolución | |---|---|---| | Mensaje de error u síntoma observable] | Por qué sucede esto] | Solución exacta o próximo paso] | | Otro síntoma] | Causa] | Resolución] |
Si este runbook no resuelve el problema:
| Condición | A Quién Contactar | Cómo | |---|---|---| | p. ej. DB no disponible después de 10 min] | DBA on-call] | Política PagerDuty: db-oncall] | | p. ej. Proveedor de pagos sin respuesta] | Contacto del proveedor] | Contacto en 1Password: vendor-escalation] |
Siempre actualiza la línea de tiempo del incidente en herramienta] antes de escalar.
Después de completar el runbook:
#ops-live con el resultado| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-15 | fail→pass | 24,560 | 31,207 | +27% | 1 | 1 | 0% | 3,557 | 6,233 | +75% | 0 | 0 | — |
case-01 | fail→pass | 33,336 | 41,697 | +25% | 1 | 1 | 0% | 5,050 | 7,334 | +45% | 0 | 0 | — |
case-02 | fail→fail | 36,272 | 52,394 | +44% | 1 | 1 | 0% | 6,439 | 7,726 | +20% | 0 | 0 | — |
case-03 | fail→pass | 47,799 | 47,416 | -1% | 1 | 1 | 0% | 8,303 | 8,218 | -1% | 0 | 0 | — |
case-04 | pass→pass | 29,353 | 32,064 | +9% | 1 | 1 | 0% | 3,738 | 5,212 | +39% | 0 | 0 | — |
case-05 | pass→fail | 40,451 | 53,208 | +32% | 1 | 1 | 0% | 5,578 | 9,952 | +78% | 0 | 0 | — |
case-06 | pass→pass | 26,534 | 27,498 | +4% | 1 | 1 | 0% | 3,055 | 5,478 | +79% | 0 | 0 | — |
case-07 | fail→pass | 40,926 | 31,561 | -23% | 1 | 1 | 0% | 5,384 | 7,925 | +47% | 0 | 0 | — |
case-08 | fail→fail | 42,283 | 30,754 | -27% | 1 | 1 | 0% | 6,289 | 5,561 | -12% | 0 | 0 | — |
case-09 | pass→fail | 54,170 | 126,836 | +134% | 1 | 1 | 0% | 5,427 | 6,565 | +21% | 0 | 0 | — |
case-10 | pass→pass | 38,414 | 36,935 | -4% | 1 | 1 | 0% | 4,890 | 6,147 | +26% | 0 | 0 | — |
case-11 | pass→pass | 61,680 | 43,734 | -29% | 1 | 1 | 0% | 5,719 | 7,641 | +34% | 0 | 0 | — |
case-12 | pass→fail | 40,892 | 32,493 | -21% | 1 | 1 | 0% | 5,316 | 6,780 | +28% | 0 | 0 | — |
case-13 | fail→pass | 31,406 | 36,556 | +16% | 1 | 1 | 0% | 4,450 | 6,468 | +45% | 0 | 0 | — |
case-14 | pass→fail | 26,568 | 47,303 | +78% | 1 | 1 | 0% | 3,820 | 9,954 | +161% | 0 | 0 | — |
case-16 | fail→fail | 20,510 | 25,861 | +26% | 1 | 1 | 0% | 3,342 | 5,482 | +64% | 0 | 0 | — |
case-17 | pass→fail | 29,233 | 43,127 | +48% | 1 | 1 | 0% | 3,757 | 7,219 | +92% | 0 | 0 | — |
case-18 | fail→fail | 42,601 | 28,146 | -34% | 1 | 1 | 0% | 3,402 | 5,781 | +70% | 0 | 0 | — |
case-19 | pass→pass | 27,027 | 33,593 | +24% | 1 | 1 | 0% | 4,008 | 6,729 | +68% | 0 | 0 | — |
case-20 | pass→pass | 41,962 | 53,910 | +28% | 1 | 1 | 0% | 5,408 | 7,346 | +36% | 0 | 0 | — |
case-21 | pass→pass | 27,492 | 28,572 | +4% | 1 | 1 | 0% | 3,794 | 6,793 | +79% | 0 | 0 | — |
case-22 | pass→fail | 27,041 | 36,231 | +34% | 1 | 1 | 0% | 3,476 | 6,208 | +79% | 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 -8 percentage points is the difference between those two pass rates over the 22 comparable cases. 7 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.