spec-driven-development

SkillDev tools

Usar cuando se escribe, valida o implementa una spec ejecutable SDD.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the spec-driven-development skill

What this skill tells your AI

The instructions your AI receives, as published by gonzalezpazmonica/savia in .claude/skills/spec-driven-development/SKILL.md and read by ahel’s review.

Subagent Scope Guard

If you were dispatched as a subagent to execute a specific delegated task, skip this skill's full orchestration workflow. Execute only the assigned task, report result (DONE / DONE_WITH_CONCERNS / BLOCKED), and return. This guard prevents runaway skill activation in nested agent contexts.

Skill: Spec-Driven Development (SDD)

Transforma Tasks de Azure DevOps en Specs ejecutables por un Developer humano o un agente Claude.

Prerequisitos: ../azure-devops-queries/SKILL.md, ../pbi-decomposition/SKILL.md


Decision Checklist

  1. Does the spec have all interfaces, types, and edge cases defined? -> If NO: return to architect/spec-writer
  2. Are acceptance criteria measurable (Given/When/Then with data)? -> If NO: return to business-analyst
  3. Is there an existing code pattern in the project to follow? -> If YES: reference as exemplar in spec
  4. Does this touch auth, payments, PII, or public APIs? -> If YES: mandate security-review first
  5. Can an agent implement without asking questions? -> If NO: developer_type = human; If YES: agent

Abort Conditions

  • Spec has TODO/TBD placeholders -> incomplete, return to spec-writer
  • No test strategy defined -> return to test-engineer for test plan

Concepto Central

PBI → Tasks → Specs (SDD) → Implementación (Human | Agent) → Code Review → Done

Un Developer puede ser:

TipoCuándo usar
humanLógica compleja, decisiones arquitectónicas, ambigüedad alta
agent-singleTasks bien definidas, patrones repetitivos, boilerplate
agent-teamTasks grandes (>6h) que benefician de paralelización

Fase 1 — Determinar Developer Type

Factores que favorecen agente: patrón claro y repetible · output determinístico (tests, DTOs, validators) · ejemplos en el código · reglas de negocio completas.

Factores que favorecen humano: lógica de dominio novedosa · trade-offs arquitectónicos · sistemas externos sin documentación · criterios incompletos · E1 (Code Review) → siempre humano.


Fase 2 — Generar Spec

2.1 Obtener información

curl -s -u ":$PAT" "$AZURE_DEVOPS_ORG_URL/{proyecto}/_apis/wit/workitems/{id}?api-version=7.1" | jq .

2.2 Inspeccionar código existente

find src -name "*{patrón}*" | head -5

2.3 Construir Spec

Guardar en: projects/{proyecto}/specs/{sprint}/AB{id}-{tipo}-{desc}.spec.md Usar plantilla: references/spec-template.md

2.4 Criterios de calidad

Una Spec es ejecutable cuando:

  • Contrato (interface) definido exactamente
  • Tipos de entrada/salida definidos
  • Reglas de negocio inequívocas
  • Test scenarios cubren casos normales y edge
  • Ficheros a crear/modificar listados
  • Ejemplos de código similar del proyecto
  • Criterios de aceptación verificables

Si NO cumple → developer_type: human

2.4b Aprobación del Spec (answer ≠ approval)

Una respuesta a una pregunta NO es aprobación — es un INPUT que cambia el spec; re-muestra y pregunta otra vez. Spec append-only (never drift) + path absoluto. Detalle: REFERENCE.md.

2.5 Agent-Note del análisis

Escribir: projects/{proyecto}/agent-notes/{ticket}-legacy-analysis-{fecha}.md


Fase 2.5 — Security Review Pre-Implementación

Ejecutar /security-review {spec}security-guardian revisa contra OWASP Top 10. Si issues 🔴 → corregir spec antes de implementar. Obligatorio para: auth, pagos, datos personales, APIs públicas, infraestructura.


Fase 2.6 — TDD Gate: Tests Antes de Implementar

  1. test-engineer escribe tests que fallan (Red)
  2. Produce: projects/{proyecto}/agent-notes/{ticket}-test-strategy-{fecha}.md
  3. GATE: developer NO puede editar código sin tests existentes

Fase 3 — Ejecutar con Agente Claude

Detalles: references/agent-invocation.md (contexto, prompts, logging, agent-notes)


Fases 4-5 — Review, Metricas e Iteracion

Detalles: references/review-metrics.md (review checklist, Azure DevOps update, metricas SDD, mejora continua)


Delta Specs (cambios incrementales)

Formato delta (ADDED/MODIFIED/REMOVED) en lugar de reescribir. Consolidar con /spec-verify al cerrar sprint. Detalle: @references/compliance-matrix.md

Referencias

Templates: references/spec-template.md · references/layer-assignment-matrix.md · references/compliance-matrix.md | Execution: references/agent-invocation.md · references/review-metrics.md | Comandos: /spec-generate, /spec-implement, /spec-review, /spec-verify

Anti-patterns

Ver satélite: REFERENCE.md — anti-patterns de AC no verificables y merge sin aprobación humana (SE-210).

Signals

GitHub stars
50
Forks
12
Last commit
Sep 2026
Hacker News mentions
3
Advanced
Catalog kind
skill
Gateway key
spec-driven-development-2
Source
github.com/gonzalezpazmonica/savia