sistema-multiagente-sdlc 1.2.1 → 1.3.0

package/CHANGELOG.md

@@ -1,5 +1,33 @@
 # Changelog
 
+## [Unreleased]
+
+_No hay cambios pendientes._
+
+## [1.3.0] — 2026-05-23
+
+### Added
+
+- ADR `0002-codegraph-spike.md` (versión Propuesta inicial): aprobar evaluación de 7 días de [CodeGraph (colbymchenry)](https://github.com/colbymchenry/codegraph). Esta versión queda en el historial git; la versión vigente es la Aceptada del bloque "Changed" anterior.
+- ADR `0003-per-phase-model-assignment.md`: extender `templates/scripts/models.yaml` con bloque opcional `phases:` para asignar modelos distintos a fases SDD (`sdd-explore`, `sdd-design`, `sdd-implement`, `sdd-review` o F0-F17). Reduce costo en exploración manteniendo precisión en F2-F3. Inspirado por el patrón `--profile-phase` de `gentle-ai`.
+- `templates/scripts/models.yaml`: bloque `phases:` opt-in con defaults documentados (Haiku para `sdd-explore`, Opus para `sdd-design`, Sonnet para `sdd-implement`/`sdd-review`).
+- Skill `edge-case-hunter`: checklist portable para revisar entradas límite, concurrencia, fallos parciales, dependencias, autorización y volumen antes de implementación.
+- `analista-requisitos.agent.md`: protocolo de elicitación avanzada para historias ambiguas antes de pasar a diseño.
+
+### Changed
+
+- ADR `0002-codegraph-spike.md` cambia de estado **Propuesta → Aceptada**. Se reemplaza la estrategia de "spike 7 días con benchmarks sintéticos obligatorios" por **"instalar y observar"**: la adopción es coexistencia con Graphify desde el día 1, sin slice dedicado, y la verificación del ahorro de tokens claimado por CodeGraph (94 % menos tool-calls según su README) se hará por observación natural durante varias sesiones reales. El ADR ahora documenta:
+  - Reglas concretas de coexistencia (Graphify-first para razonamiento humano y `enrich-us` 4.5; CodeGraph vía MCP para queries estructurales runtime).
+  - Triggers para abrir ADR 0004 (evidencia de ahorro real, cobertura de rutas framework, o costo operativo excesivo).
+  - Implementación operativa ya ejecutada en `FacturacionDian` con scope narrow (`apps/**/src` + `packages/**/src`, `.ts/.tsx`) para evitar el OOM del index default sobre un monorepo grande.
+  - Métricas del index inicial sobre `FacturacionDian`: 658 archivos, 7.954 nodes, 14.869 edges, 14.48 MB native SQLite.
+- `scripts/validate-models-schema.mjs`: soporte para clave top-level `phases` opcional + verificación de shape `{ primary, fallback }` por fase. Bloque `phases:` declarado pero vacío produce error explícito.
+
+### Docs
+
+- README: actualizada tabla `BMAD Comparison` con realidad V6 de BMAD-METHOD (module ecosystem BMM/BMB/TEA/BMGD/CIS, Skills Architecture, Sub-Agent inclusion, scale-adaptive, Discord community) y comparación side-by-side honesta. Datos tomados del README oficial v6 de `bmad-code-org/BMAD-METHOD`.
+- README Roadmap v1.3.0: agregadas entradas `macos-latest` para `regression-install` matrix y bump de `actions/checkout@v5` + `actions/setup-node@v5` con `node-version: 24` antes del deprecation deadline de GitHub (Node 20 deprecated jun 2026, removed sep 2026).
+
 ## [1.2.1] — 2026-05-18
 
 ### UX fix — `init` sin `--target` usa cwd

package/README.md

@@ -122,21 +122,35 @@ All external installs are opt-in. Scripts default to dry-run or local-only behav
 
 ## BMAD Comparison
 
-| Feature | BMAD-METHOD | SistemaMultiagente_SDLC |
+Side-by-side de los dos frameworks. La intención no es competir sino aclarar dónde se solapan y dónde cada uno se especializa. Datos de BMAD tomados de su README oficial v6 (`bmad-code-org/BMAD-METHOD`, npm `bmad-method`).
+
+| Feature | BMAD-METHOD v6 | SistemaMultiagente_SDLC v1.3.0 |
 | --- | --- | --- |
-| AI-driven agents | 12+ personas | 8 personas plus extensible roadmap |
-| Workflows | agile | SDD waterfall by slice, agile by release |
-| Scale-adaptive | yes | reserved for v1.3.0 |
-| Party mode | yes | roundtable opt-in planned for v1.3.0 |
-| Help CLI | bmad-help | `sdlc next` planned for v1.3.0 |
-| Modules/packs | yes | packs marketplace planned for v2.0.0 |
-| Governance validators | not core | 14 validators |
-| OpenSpec/SDD | not core | integrated |
-| Readiness L1/L2/L3 | not core | integrated |
-| Multi-agent lock | not core | TTL platform-context lock |
-| Brownfield-first | no | yes |
-| Migration system | not core | backup and rollback |
-| Sanitization validators | not core | no-personal-paths and template-sanitization |
+| License | MIT | MIT |
+| Runtime requisitos | Node ≥20.12, Python ≥3.10, `uv` | Node ≥18, PowerShell (pwsh/powershell), Git |
+| Install command | `npx bmad-method install` (interactive) o `--yes --modules --tools` (CI) | `npx sistema-multiagente-sdlc init` (cwd default desde v1.2.1) |
+| Scope principal | AI-driven agile development | AI-assisted SDLC con governance enterprise y SDD |
+| Workflows | 34+ agile workflows (BMM core) | SDD waterfall por slice + agile por release (F0-F17 phases) |
+| Scale-adaptive | sí, automático (bug → enterprise) | scale hint activo desde v1.3.0 |
+| Agentes/personas | 12+ personas (PM, Architect, Dev, UX, …) | 8 personas activas + roadmap extensible |
+| Party / collaboration mode | yes (multiple personas en sesión) | roundtable opt-in planned v1.3.0 |
+| Help CLI / next-step coach | `bmad-help` skill | `sdlc doctor` (state checks); `sdlc next` planned v1.3.0 |
+| Modules / ecosystem | BMM (core) + BMB (builder) + TEA (test architect) + BMGD (game dev) + CIS (creative) | mode-based (`greenfield` / `legacy`) + extensible packs planned v2.0.0 |
+| Skills architecture | sí (V6 + Sub-Agent inclusion + Cross-Platform Agent Team) | skills mirroring across `.claude/`, `.agents/`, `.windsurf/` (`bootstrap-agent-skills.ps1`) |
+| Custom agent/workflow builder | BMad Builder v1 | personas `.agent.md` + validators (`validate-agent-persona-schema`) |
+| Dev Loop automation | en roadmap V6 | `phase-graph.yaml` + rework label-driven + lock TTL |
+| Brownfield-first | no | sí (legacy mode con research obligatorio antes de proposal) |
+| Governance validators | not core | 14 validators (config, personal-paths, template-sanitization, manifest-integrity, governance-precedence, …) |
+| OpenSpec / SDD | not core | integrated (capacidades canónicas en `openspec/specs/`) |
+| Readiness L1/L2/L3 + matriz NFR | not core | integrated (`business-production-readiness` spec) |
+| Migration system + rollback | not core | backup automático + `sdlc upgrade --to-version` + `sdlc rollback --to <id>` |
+| Multi-agent lock | not core | TTL `platform-context.json` lock |
+| Sanitization de paths/templates | not core | `validate:no-personal-paths` + `validate:template-sanitization` |
+| Provenance (SLSA) | n/d explícito | sí, SLSA v1 + signatures vía OIDC GitHub (workflow `publish.yml`) |
+| Community | Discord abierto, YouTube, X | GitHub Issues + Discussions (Discord no necesario) |
+| Trademark | BMad / BMAD-METHOD trademarks of BMad Code, LLC | sin restricción explícita más allá de MIT |
+
+Lectura corta: BMAD lidera en agile breadth y community (12+ personas, 34+ workflows, 5 módulos, Discord activo, Skills Architecture V6). SistemaMultiagente_SDLC lidera en governance + brownfield + SDD + validators (14) + migration system + readiness L1/L2/L3 + sanitization. Ambos pueden coexistir: BMAD orquesta; SistemaMultiagente_SDLC orquesta **y verifica**.
 
 ## Roadmap
 
@@ -148,6 +162,8 @@ v1.3.0:
 - calibration extensions
 - roundtable opt-in
 - docs site
+- regression-install matrix: agregar `macos-latest` (triple coverage ubuntu + windows + macos)
+- bump `actions/checkout@v5` + `actions/setup-node@v5` con `node-version: 24`; deadline GitHub: Node 20 deprecated jun 2026, removed sep 2026
 
 v2.0.0:
 

package/migrations/1.3.0/up.mjs

@@ -0,0 +1,14 @@
+export function up(files = {}) {
+  const extra = {
+    ".sdlc/migrations/1.3.0-applied.txt": "Migration 1.3.0 applied by SistemaMultiagente_SDLC.\ngenerated-by-sdlc\n"
+  };
+
+  const configPath = ".sdlc/config.json";
+  if (typeof files[configPath] === "string") {
+    const config = JSON.parse(files[configPath]);
+    config.frameworkVersion = "1.3.0";
+    extra[configPath] = `${JSON.stringify(config, null, 2)}\n`;
+  }
+
+  return extra;
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "sistema-multiagente-sdlc",
-  "version": "1.2.1",
+  "version": "1.3.0",
   "description": "Framework reusable para instalar un SDLC multiagente gobernado en repos greenfield o legacy.",
   "type": "module",
   "bin": {

package/schemas/sdlc.config.schema.json

@@ -21,7 +21,7 @@
       "type": "string",
       "enum": ["bug", "feature", "epic", "platform"],
       "default": "feature",
-      "description": "Reserved for v1.3.0 adaptive scale. Not consumed in v1.2.0."
+      "description": "Scale hint used by v1.3.0+ for adaptive planning and agent routing."
     },
     "stack": {
       "type": "object",

package/scripts/validate-models-schema.mjs

@@ -25,7 +25,31 @@ for (const platform of ["claude_code", "codex", "copilot", "windsurf"]) {
   requirePattern(new RegExp(`^  ${platform}:\\s*\\n    default:\\s*\\S+`, "m"), `platform ${platform} requires default`);
 }
 
-const allowedTopLevel = new Set(["version", "updated", "roles", "platforms", "fallbacks"]);
+// Optional `phases:` block (v1.3.0+). When present, each phase entry must
+// declare primary and fallback model identifiers, same shape as `roles`.
+if (/^phases:\s*$/m.test(content)) {
+  const phasesBlockRegex = /^phases:\s*$([\s\S]*?)(?=^[A-Za-z0-9_-]+:|\Z)/m;
+  const phasesBlockMatch = phasesBlockRegex.exec(content);
+  if (phasesBlockMatch) {
+    const phasesBody = phasesBlockMatch[1];
+    const phaseEntryRegex = /^  ([A-Za-z0-9_-]+):\s*$/gm;
+    let phaseMatch;
+    let phasesFound = 0;
+    while ((phaseMatch = phaseEntryRegex.exec(phasesBody)) !== null) {
+      phasesFound += 1;
+      const phaseName = phaseMatch[1];
+      const phaseShape = new RegExp(`^  ${phaseName}:\\s*\\n    primary:\\s*\\S+\\n    fallback:\\s*\\S+`, "m");
+      if (!phaseShape.test(phasesBody)) {
+        errors.push(`phase ${phaseName} requires primary and fallback`);
+      }
+    }
+    if (phasesFound === 0) {
+      errors.push("phases block declared but contains no phase entries");
+    }
+  }
+}
+
+const allowedTopLevel = new Set(["version", "updated", "roles", "platforms", "phases", "fallbacks"]);
 for (const line of content.split("\n")) {
   const match = /^([A-Za-z0-9_-]+):/.exec(line);
   if (match && !allowedTopLevel.has(match[1])) {

package/src/cli.js

@@ -392,7 +392,7 @@ function collectDoctorEnhancements(target, config) {
   }
 
   if (!config || !config.scale) {
-    findings.push({ level: "error", code: "scale-missing", message: "config.scale is required in v1.2.0" });
+    findings.push({ level: "error", code: "scale-missing", message: "config.scale is required in v1.3.0" });
   } else {
     findings.push({ level: "info", code: "scale-present", message: `scale=${config.scale}` });
   }

package/src/migrations.js

@@ -1,11 +1,13 @@
 import { up as up_1_0_1 } from "../migrations/1.0.1/up.mjs";
 import { up as up_1_1_0 } from "../migrations/1.1.0/up.mjs";
 import { up as up_1_2_0 } from "../migrations/1.2.0/up.mjs";
+import { up as up_1_3_0 } from "../migrations/1.3.0/up.mjs";
 
 const REGISTRY = [
   { version: "1.0.1", up: up_1_0_1 },
   { version: "1.1.0", up: up_1_1_0 },
-  { version: "1.2.0", up: up_1_2_0 }
+  { version: "1.2.0", up: up_1_2_0 },
+  { version: "1.3.0", up: up_1_3_0 }
 ];
 
 function semverTuple(v) {

package/src/render.js

@@ -2,7 +2,7 @@ import path from "node:path";
 import { renderTemplates } from "./template-loader.js";
 export { validateConfigShape } from "./config-validator.js";
 
-export const FRAMEWORK_VERSION = "1.2.0";
+export const FRAMEWORK_VERSION = "1.3.0";
 export const SCHEMA_VERSION = 1;
 export const SUPPORTED_MODES = new Set(["greenfield", "legacy"]);
 

package/templates/.github/agents/analista-requisitos.agent.md

@@ -23,3 +23,34 @@ Owns functional definition before planning and implementation begin.
 - Skip human validation for functional scope.
 - Treat ambiguous stakeholder language as executable specification.
 - Decide architecture ownership without involving `arquitecto`.
+
+## Advanced Elicitation
+
+When a story is ambiguous, incomplete or contradictory, apply this protocol before producing the functional draft.
+
+### Clarification Dimensions
+
+Ask at least one question for every undefined dimension:
+
+| Dimension | Example question |
+|---|---|
+| Functional | What must the user be able to do at the end? What works today and must not change? |
+| NFR | Is there a response-time, concurrency or data-size constraint? |
+| UX | Is there visible UI? Where does it happen? How does the user know it finished? |
+| Integration | Which external systems participate? What happens if one fails? |
+
+### Minimum Acceptance Criteria
+
+Do not move to design until these exist:
+
+1. Actor + action + expected result.
+2. Happy path + one unhappy path.
+3. Functional Definition of Done.
+
+### Escalation
+
+If the stakeholder cannot answer in one elicitation session:
+
+1. Mark gaps in the draft with `[GAP: undefined]`.
+2. Propose a 48-hour deadline in the validation block.
+3. If still undefined after 48 hours, escalate to the human business owner and do not move to design.

package/templates/.github/skills/edge-case-hunter/SKILL.md

@@ -0,0 +1,71 @@
+---
+name: edge-case-hunter
+description: "Structured edge-case checklist for F3/F4 before implementation. Adapted from BMAD-METHOD patterns without installing the full framework."
+---
+
+# Edge Case Hunter
+
+## Goal
+
+Produce an edge-case inventory for a capability before implementation starts. The goal is not to block delivery; it is to force hard cases into design time instead of QA or production.
+
+## When To Use
+
+- **F3 Design** — before closing design and moving to implementation.
+- **F4 Validation** — pre-merge review when design skipped edge-case review.
+
+## Checklist
+
+Answer every dimension. A "not applicable" answer must include a one-line justification.
+
+### 1. Invalid Or Boundary Input
+
+- [ ] Empty, null or undefined input.
+- [ ] Zero-size input: empty array, empty string, object without properties.
+- [ ] Valid min and max boundary.
+- [ ] Just outside valid boundary.
+- [ ] Wrong types.
+
+### 2. Concurrency And Shared State
+
+- [ ] Two users execute the same operation on the same resource.
+- [ ] Read/write race condition.
+- [ ] Retry after partial state change.
+
+### 3. Partial Failure And Recovery
+
+- [ ] Step N fails after step N-1 already changed state.
+- [ ] Rollback or explicit partial-state handling.
+- [ ] Error message names the failing step enough for recovery.
+
+### 4. External Dependencies
+
+- [ ] External service timeout.
+- [ ] Unexpected response shape.
+- [ ] Retry idempotency.
+
+### 5. Authorization And Privacy
+
+- [ ] User changes an ID in the URL or payload.
+- [ ] Logs or errors expose sensitive data.
+- [ ] Permission validation happens on the server, not only in UI.
+
+### 6. Volume And Performance
+
+- [ ] Large result set.
+- [ ] Required indexes for queries.
+- [ ] Explicit response limits.
+
+## Expected Output
+
+Add an `### Edge cases` section to the design or local issue with:
+
+- Cases that pass.
+- Cases that create implementation tasks.
+- Cases that do not apply, with justification.
+
+## Limits
+
+- Do not invent unrelated cases.
+- If investigation takes more than 30 minutes, create a spike.
+- This complements QA; it does not replace it.

package/templates/docs/agents/catalogo-comandos-proyecto.md

@@ -8,7 +8,7 @@
 | `sdlc install --mode legacy` | instala gobierno en proyecto brownfield |
 | `sdlc doctor --json` | valida estado instalado |
 | `sdlc diff --json` | detecta drift contra templates |
-| `sdlc upgrade --to-version 1.2.0` | aplica migraciones |
+| `sdlc upgrade --to-version 1.3.0` | aplica migraciones |
 | `sdlc rollback --to <backup>` | restaura backup |
 | `sdlc prune-backups --keep 5` | limpia backups antiguos |
 

package/templates/manifest.yaml

@@ -172,6 +172,8 @@ templates:
     target: .claude/skills/enrich-us/SKILL.md
   - source: .github/skills/enrich-us/SKILL.md
     target: .github/skills/enrich-us/SKILL.md
+  - source: .github/skills/edge-case-hunter/SKILL.md
+    target: .github/skills/edge-case-hunter/SKILL.md
   - source: .claude/skills/commit/SKILL.md
     target: .claude/skills/commit/SKILL.md
   - source: .github/skills/commit/SKILL.md

package/templates/scripts/agent-skills.manifest.json

@@ -14,6 +14,7 @@
     "commit",
     "contexto-proyecto",
     "documentacion-viva",
+    "edge-case-hunter",
     "enrich-us",
     "openspec-apply",
     "openspec-archive",

package/templates/scripts/models.yaml

@@ -1,5 +1,5 @@
 version: 1
-updated: "2026-05-18"
+updated: "2026-05-19"
 
 roles:
   orquestador:
@@ -25,6 +25,36 @@ platforms:
   windsurf:
     default: claude-sonnet-4-6
 
+# Per-phase model assignment (opt-in, v1.3.0+).
+# Asignar modelos distintos a cada fase del flujo SDD para optimizar costo sin
+# perder calidad en pasos críticos. Cuando una fase no aparezca aquí, se usa el
+# rol equivalente (ver bloque `roles` arriba). Esto NO reemplaza `roles`/`platforms`;
+# es una capa de override más fina.
+#
+# Convenciones:
+# - Claves: nombres canónicos de fase F0-F17 (ver docs/agents/presentacion-sistema-multiagente-sdlc.md)
+#   o pseudo-fases SDD (`sdd-explore`, `sdd-design`, `sdd-implement`, `sdd-review`).
+# - Valores: { primary, fallback }, mismo schema que `roles`.
+# - Modelos `cheap:*` o `:free` quedan permitidos para exploración/triage donde la
+#   precisión no es crítica. Ver ADR 0003 para criterios.
+phases:
+  # F0 triage / exploración rápida → modelo barato
+  sdd-explore:
+    primary: claude-haiku-4-5-20251001
+    fallback: claude-sonnet-4-6
+  # F2-F3 análisis y diseño → modelo más capaz
+  sdd-design:
+    primary: claude-opus-4-7
+    fallback: claude-sonnet-4-6
+  # F5+ implementación → modelo intermedio
+  sdd-implement:
+    primary: claude-sonnet-4-6
+    fallback: claude-haiku-4-5-20251001
+  # F6+ verificación / review → modelo intermedio con fallback a haiku
+  sdd-review:
+    primary: claude-sonnet-4-6
+    fallback: claude-haiku-4-5-20251001
+
 fallbacks:
   - claude-sonnet-4-6
   - claude-haiku-4-5-20251001