@atlashub/smartstack-cli 3.39.0 → 3.40.0

package/templates/skills/business-analyse/steps/step-03d-validate.md

@@ -1,783 +1,783 @@
----
-name: step-03d-validate
-description: Per-module validation, write feature.json, incremental HTML deploy, loop decision
-model: sonnet
-next_step: steps/step-03a1-setup.md OR steps/step-04a-collect.md (conditional)
----
-
-> **Context files:** `_shared.md` | `_module-loop.md`
-
-# Step 3d: Validation & Loop
-
-## MANDATORY EXECUTION RULES
-
-- ALWAYS use ULTRATHINK mode
-- This step VALIDATES the specification from step-03c, writes it to feature.json, and decides loop continuation
-- ALWAYS verify specification completeness before writing
-- ALL communication in `{language}`
-
-## MODE DETECTION (inherited from step-03a1)
-
-> **CRITICAL: Re-check your execution mode before proceeding.**
-
-**IF you are running as a TEAM AGENT** (your prompt contains `PROPOSE & REVIEW` or `team-lead` as recipient):
-  → **NEVER** use `AskUserQuestion` in ANY section below (sections 9d, 10, 12)
-  → Section 9d: if validation fails, AUTO-CORRECT silently (no user options)
-  → Section 12: **SKIP ENTIRELY** — go directly to section 12-bis (Agent Mode)
-  → After writing feature.json, send `PROPOSAL_READY` to team lead via SendMessage
-  → NEVER present options/menus to the user — you are an autonomous agent
-
-**IF you are running in the MAIN CONVERSATION** (classic inline mode):
-  → Normal interactive mode — use `AskUserQuestion` as documented
-  → Section 12-bis does NOT apply to you
-- **ID NAMING RULE (MANDATORY, NO EXCEPTION):**
-  All IDs MUST include a module prefix to guarantee application-wide uniqueness.
-  The prefix is derived from the module code initials (2-4 chars):
-    UserManagement → UM | VehicleManagement → VM | PartsInventory → PI
-    RepairManagement → RM | MaintenanceSchedule → MS | DataSync → DS
-    Notifications → NT | Dashboard → DB | Orders → OR | Customers → CU
-
-  NEVER use bare IDs (UC-001, BR-VAL-001) in multi-module mode.
-
-## YOUR TASK
-
-Validate the module specification for completeness and consistency, write to feature.json, deploy incremental HTML, and decide whether to loop to the next module or proceed to consolidation.
-
----
-
-### 9. Per-Module Validation
-
-#### 9a. Completeness Checks
-
-> **CRITICAL:** Checks MUST count ACTUAL elements in arrays, NOT declarative values.
-> A check that reports PASS when the array is empty is a LIE and BLOCKS downstream quality.
-
-| Section | Minimum | How to verify | Status |
-|---------|---------|---------------|--------|
-| actors | 2 | `specification.actors.length >= 2` | PASS/FAIL |
-| useCases | 2 | `specification.useCases.length >= 2` | PASS/FAIL |
-| functionalRequirements | 4 | `specification.functionalRequirements.length >= 4` | PASS/FAIL |
-| permissionMatrix | 1 resource × 2 roles | `specification.permissionMatrix.permissions.length >= 1 && specification.permissionMatrix.roleAssignments.length >= 2` | PASS/FAIL |
-| entities | 1 | `analysis.entities.length >= 1` | PASS/FAIL |
-| entitySchemaFormat | attributes[] not fields[] (BLOCKING) | `analysis.entities.every(e => e.attributes?.length > 0)` | PASS/FAIL |
-| entityAttributeTypes | ALL attributes have `type` field (BLOCKING) | `analysis.entities.every(e => e.attributes.every(a => a.type))` — **NOTE:** Entities in analysis phase may NOT have `type`. Step-03c auto-fix MUST have run first. If types are missing, RE-RUN step-03c auto-fix algorithm. | PASS/FAIL |
-| navigation | entries[] present (BLOCKING) | `specification.navigation?.entries?.length >= 1` | PASS/FAIL |
-| dataLifecycle | present in analysis (BLOCKING) | `analysis.dataLifecycle !== undefined` | PASS/FAIL |
-| wireframes | 1 per section (BLOCKING) | `(specification.uiWireframes \|\| specification.wireframes \|\| []).length >= (specification.sections \|\| []).length` (count REAL elements, check BOTH key names) | PASS/FAIL |
-| wireframeSchema | All required fields present (BLOCKING) | `(specification.uiWireframes \|\| specification.wireframes \|\| []).every(w => (w.screen \|\| w.title) && w.section && (w.mockup \|\| w.ascii \|\| w.content))` | PASS/FAIL |
-| sections | 1 (BLOCKING) | `specification.sections.length >= 1` (EVERY module needs at least 1 section) | PASS/FAIL |
-| gherkinScenarios | 1 array entry | `Array.isArray(specification.gherkinScenarios) && specification.gherkinScenarios.length >= 1` | PASS/FAIL |
-| gherkinFormat | Array not object (BLOCKING) | `Array.isArray(specification.gherkinScenarios)` (NOT a single object) | PASS/FAIL |
-| validations | 1 | `specification.validations.length >= 1` | PASS/FAIL |
-| validationFormat | rules[] array (BLOCKING) | `specification.validations.every(v => Array.isArray(v.rules))` (NOT singular `rule`) | PASS/FAIL |
-| messages | 4 | `specification.messages.length >= 4` | PASS/FAIL |
-| messageFormat | `message` field present (BLOCKING) | `specification.messages.every(m => m.message)` | PASS/FAIL |
-| lifeCycles | 1 (if entity has status) | `specification.lifeCycles.length >= 1` (if any entity has status/state field) | PASS/FAIL |
-| seedDataCore | 9 arrays present with content | See detailed check below | PASS/FAIL (BLOCKING) |
-| apiEndpoints | 1 | `specification.apiEndpoints.length >= 1` | PASS/FAIL |
-| i18nKeys | present | `specification.i18nKeys !== undefined && specification.i18nKeys !== null` | PASS/FAIL |
-| navigationIcons | non-null | `specification.seedDataCore.navigationModules.every(m => m.icon !== null)` | PASS/FAIL |
-
-**seedDataCore detailed check (BLOCKING):**
-```javascript
-const sdc = specification.seedDataCore;
-const currentModuleIndex = metadata.workflow?.currentModuleIndex || 0;
-const checks = [
-  // navigationApplications and applicationRoles: required for first module (index 0), can be empty for subsequent modules
-  { key: "navigationApplications", actual: sdc.navigationApplications?.length || 0, min: currentModuleIndex === 0 ? 1 : 0 },
-  { key: "applicationRoles",      actual: sdc.applicationRoles?.length || 0,       min: currentModuleIndex === 0 ? 1 : 0 },
-  { key: "navigationModules",      actual: sdc.navigationModules?.length || 0,      min: 1 },
-  { key: "navigationSections",     actual: sdc.navigationSections?.length || 0,     min: 1 },  // EVERY module needs ≥1 section
-  { key: "navigationResources",    actual: sdc.navigationResources?.length || 0,    min: 1 },
-  { key: "navigationTranslations", actual: sdc.navigationTranslations?.length || 0, min: 2 },  // min fr+en
-  { key: "permissions",            actual: sdc.permissions?.length || 0,            min: 1 },
-  { key: "rolePermissions",        actual: sdc.rolePermissions?.length || 0,        min: 1 },
-  { key: "permissionConstants",    actual: sdc.permissionConstants?.length || 0,    min: 1 }
-];
-const failures = checks.filter(c => c.actual < c.min);
-IF failures.length > 0:
-  BLOCKING ERROR: "seedDataCore incomplete — empty arrays: {failures.map(f => f.key).join(', ')}"
-  → Fix: Ensure specification.sections[] has ≥1 entry, then re-run 8f-bis transform
-```
-
-#### 9b. Consistency Checks
-
-- Every UC has ≥1 linked FR
-- Every FR has ≥1 linked BR
-- All BR references exist in analysis.businessRules
-- All actors appear in permissionMatrix
-- Permission paths use correct format: module-level `{app}.{module}.{action}` or section-level `{app}.{module}.{section}.{action}`
-- rolePermissions paths match permissions paths
-- API routes use consistent prefix
-
-#### 9c. Convention Checks
-
-- UC-NNN format
-- FR-NNN format
-- BR-{CATEGORY}-NNN format
-- Entity names PascalCase
-- Field names camelCase
-- Entity attribute format: `attributes[]` with {name, description}, NOT `fields[]` with {name, type} — entities must NOT have tableName or primaryKey
-- Wireframe structure: `screen` (not `name`/`title`), `mockup` (not `ascii`/`content`), `componentMapping` is array of {wireframeElement, reactComponent} (not plain key-value object), `layout` is object with regions (not string)
-- Wireframe field naming: uses `screen` (not `title`), `mockup` (not `ascii`) — auto-fix if wrong (see 9c-fix below)
-- Permission paths dot-separated lowercase
-
-#### 9c-fix. Auto-Fix Wireframe Field Names (MANDATORY before writing)
-
-> **CRITICAL:** The agent may use non-canonical field names (`title`, `ascii`, `name`, `content`).
-> These MUST be normalized to canonical names before writing to feature.json.
-
-```javascript
-// AUTO-FIX: Normalize wireframe field names before writing
-const wireframes = specification.uiWireframes || specification.wireframes || [];
-for (const wf of wireframes) {
-  if (wf.title && !wf.screen) { wf.screen = wf.title; delete wf.title; }
-  if (wf.ascii && !wf.mockup) { wf.mockup = wf.ascii; delete wf.ascii; }
-  if (wf.content && !wf.mockup) { wf.mockup = wf.content; delete wf.content; }
-  if (wf.name && !wf.screen) { wf.screen = wf.name; delete wf.name; }
-}
-// Store normalized wireframes back under canonical key
-specification.uiWireframes = wireframes;
-```
-
-#### 9c-fix-entities. Auto-Fix Entity Attribute Types (MANDATORY before writing)
-
-> **DEFENSE-IN-DEPTH:** step-03c POST-CHECK should catch attributes without `type`,
-> but if any slip through to step-03d, auto-fix them here before writing to feature.json.
-> Uses the same type inference algorithm as step-03c POST-CHECK.
-
-```javascript
-// AUTO-FIX: Ensure all entity attributes have structured type
-const entities = analysis.entities || [];
-let entityAutoFixCount = 0;
-for (const entity of entities) {
-  for (const attr of entity.attributes || []) {
-    if (!attr.type) {
-      // Infer type from attribute name pattern
-      if (attr.validation?.match(/max\s*\d+/i) || attr.name.match(/name|title|code|description|label|email|phone|address/i)) {
-        attr.type = "string";
-        const maxMatch = attr.validation?.match(/max\s*(\d+)/i);
-        if (maxMatch) attr.maxLength = parseInt(maxMatch[1]);
-      } else if (attr.name.match(/id$/i)) {
-        attr.type = "Guid";
-      } else if (attr.name.match(/date|At$/i)) {
-        attr.type = "DateTime";
-      } else if (attr.name.match(/is[A-Z]|has[A-Z]|active|enabled/)) {
-        attr.type = "bool";
-      } else if (attr.name.match(/amount|salary|rate|price|total/i)) {
-        attr.type = "decimal";
-      } else if (attr.name.match(/count|number|order|sort|index/i)) {
-        attr.type = "int";
-      } else {
-        attr.type = "string"; // safe default
-      }
-      // Normalize free-text maxLength
-      if (typeof attr.validation === 'string' && !attr.maxLength) {
-        const m = attr.validation.match(/max\s*(\d+)/i);
-        if (m) attr.maxLength = parseInt(m[1]);
-      }
-      if (attr.required === undefined) attr.required = true;
-      entityAutoFixCount++;
-    }
-  }
-}
-if (entityAutoFixCount > 0) {
-  console.warn(`DEFENSE-IN-DEPTH: auto-fixed ${entityAutoFixCount} attributes without type in step-03d`);
-}
-```
-
-#### 9d. Decision
-
-> **TEAM AGENT MODE:** If running as team agent, SKIP AskUserQuestion. On FAIL → auto-correct silently. On PASS → proceed directly to section 11 (write). NEVER present options to the user.
-
-**INLINE MODE ONLY (main conversation):**
-
-IF validation PASS:
-  Display summary, ask client for confirmation
-
-IF validation FAIL:
-  Display errors, offer:
-  ```
-  options:
-    - label: "Auto-corriger"
-      description: "Corriger automatiquement les erreurs détectées"
-    - label: "Réviser manuellement"
-      description: "Revenir aux sections problématiques"
-    - label: "Forcer la validation"
-      description: "Accepter malgré les erreurs (risqué)"
-  ```
-
-#### 9e. Write Module Validation Section (MANDATORY)
-
-> **The module feature.json MUST have a `validation` section recording the results of steps 9a-9d.**
-
-```
-ba-writer.enrichSection({
-  featureId: {module_feature_id},
-  section: "validation",
-  data: {
-    validatedAt: "{ISO timestamp}",
-    completenessChecks: [
-      { "section": "useCases", "count": {count}, "minimum": 2, "status": "PASS|FAIL" },
-      { "section": "functionalRequirements", "count": {count}, "minimum": 4, "status": "PASS|FAIL" },
-      { "section": "wireframes", "count": {count}, "minimum": {section_count}, "status": "PASS|FAIL" },
-      { "section": "seedDataCore", "count": 7, "minimum": 7, "status": "PASS|FAIL" }
-    ],
-    consistencyChecks: [
-      { "check": "UC-FR linkage", "passed": {n}, "warnings": {n}, "errors": {n}, "status": "PASS|FAIL" },
-      { "check": "BR references valid", "passed": {n}, "warnings": {n}, "errors": {n}, "status": "PASS|FAIL" },
-      { "check": "Permission path format", "passed": {n}, "warnings": {n}, "errors": {n}, "status": "PASS|FAIL" }
-    ],
-    conventionChecks: [
-      { "check": "ID naming with module prefix", "status": "PASS|FAIL", "details": "..." },
-      { "check": "Entity names PascalCase", "status": "PASS|FAIL", "details": "..." }
-    ],
-    semanticChecks: [
-      { "check": "orphan-permissions", "module": "{currentModule}", "status": "PASS|WARNING|ERROR", "details": "...", "autoFixed": false },
-      { "check": "uc-sans-fr", "module": "{currentModule}", "status": "PASS|WARNING|ERROR", "details": "...", "autoFixed": false },
-      { "check": "missing-wireframes", "module": "{currentModule}", "status": "PASS|WARNING|ERROR", "details": "...", "autoFixed": false }
-    ],
-    warnings: ["List of non-blocking warnings"],
-    decision: {
-      "approved": true,
-      "reason": "All checks passed",
-      "approvalMode": "standard",
-      "approvedBy": "Client",
-      "approvedAt": "{ISO timestamp}"
-    }
-  }
-})
-```
-
-#### 9e-POST-CHECK: Validation Section Persistence (BLOCKING)
-
-> **CRITICAL:** The `validation` section MUST be persisted to feature.json. Without it, consolidation checks fail.
-
-```bash
-MODULE_JSON="{module_feature_json_path}"
-node -e "const d=JSON.parse(require('fs').readFileSync(process.argv[1],'utf-8'));
-if(!d.validation||!d.validation.decision) { console.error('FAIL: validation section missing or incomplete'); process.exit(1); }
-console.log('PASS: validation section present with decision');" "$MODULE_JSON"
-```
-
-IF this check FAILS → re-execute section 9e write and re-run this check.
-
-#### 9f. Module Specification Checklist (BLOCKING)
-
-> **CRITICAL:** This checklist MUST be FULLY COMPLETED before marking module status = "specified".
-
-**Execute the comprehensive validation checklist:**
-
-Run the 29-check validation process across 10 categories:
-- Data Model (4 checks) | Business Rules (3 checks) | Use Cases & FRs (4 checks)
-- Permissions (3 checks) | UI & Navigation (4 checks) | I18N & Messages (3 checks)
-- Seed Data (2 checks) | API Endpoints (2 checks) | Validations (1 check) | Gherkin (3 checks)
-
-```javascript
-const blockingFailures = checks.filter(c => c.blocking && c.status === "FAIL");
-
-IF blockingFailures.length > 0:
-  // CONDITIONAL LOAD: Only load detailed checklist on validation failure
-  Read references/validation-checklist.md
-  Display:
-    - Full validation logic for FAILED checks
-    - Exact requirements (minimums, patterns, formats)
-    - Fix recommendations for each failure
-
-  BLOCKING ERROR: Module specification incomplete
-  → Fix all failures before proceeding
-
-ELSE:
-  All blocking checks passed ✓
-  Display: "{warnings_count} warnings (non-blocking)"
-  → Proceed to mark module as "specified"
-```
-
-**Optimization:** The detailed 280-line validation-checklist.md is loaded **only when validation fails** (saves ~10,000 tokens on success path).
-
----
-
-### 9g. Anti-Premature-Completion Guard (MANDATORY)
-
-> **CRITICAL — NEVER say "the analysis is complete" or "ready for /ralph-loop" at this point.**
-> Step-03d is the END of SPECIFICATION, NOT the end of the BA workflow.
-> Steps 04a → 04b → 04c (consolidation) + 05a → 05b → 05c (handoff) are STILL REQUIRED.
-> Without handoff, /ralph-loop will generate an INCOMPLETE PRD (no frontend, no tests).
-
-**FORBIDDEN phrases in step-03d output:**
-- "L'analyse métier est complète" / "The business analysis is complete"
-- "Prêt pour /ralph-loop" / "Ready for /ralph-loop"
-- "Vous pouvez maintenant lancer /ralph-loop"
-- Any variation suggesting the BA is finished or /ralph-loop can be invoked
-
-**REQUIRED phrasing:**
-- "Module {name} spécifié. {Remaining} modules restants." (if more modules)
-- "Tous les modules spécifiés. Passage à la consolidation (étapes 04a-04c)..." (if last module)
-- NEVER mention /ralph-loop — only step-05c (ralph readiness) validates the handoff
-
-### 10. Module Summary with Roles & Permissions
-
-Display comprehensive summary:
-
-```
-═══════════════════════════════════════════════════════════
-  MODULE SPÉCIFIÉ: {currentModule} ({currentModuleIndex + 1}/{total})
-═══════════════════════════════════════════════════════════
-
-| Aspect | Valeur |
-|--------|--------|
-| Entités | {count} ({entity names}) |
-| Business Rules | {count} (VAL: X, CALC: Y, WF: Z) |
-| Use Cases | {count} (UC-001 → UC-NNN) |
-| Exigences | {count} FRs |
-| Sections | {count} ({section names}) |
-| Maquettes | {count} wireframes |
-| Permissions | {count} paths |
-| Tests Gherkin | {count} scénarios |
-| Messages | {count} |
-
-### Rôles & Permissions pour {currentModule}
-| Rôle | Permissions |
-|------|------------|
-| {App} Admin | {app}.{module}.* |
-| {App} Manager | read, create, update, assign |
-| {App} Contributor | read, create, update |
-| {App} Viewer | read |
-
-→ Validation: {PASS/FAIL}
-═══════════════════════════════════════════════════════════
-
-{IF validation PASS}:
-  → Module validé automatiquement
-  → Passage automatique au module suivant (section 12)
-
-{IF validation FAIL}:
-  → Afficher les erreurs et offrir les options de correction (section 9d)
-```
-
----
-
-### 11. Write Module Feature.json
-
-```
-// Write analysis section with ALL subsections
-ba-writer.enrichSection({
-  featureId: {module_feature_id},
-  section: "analysis",
-  data: {
-    objectives: [{id, objective, metric, target}],       // from 6a
-    entities: [{name, description, attributes, relationships}],  // from 6b — SCHEMA FORMAT
-    businessRules: [{id, name, category, statement, priority, conditions, examples, testability}],  // from 7 — SCHEMA FORMAT
-    processFlow: {entryPoints, mainFlow, decisionPoints, alternativeFlows},  // from 6c
-    dataLifecycle: {retentionPeriod, archiveStrategy, gdprCompliance, states}  // from 6d
-  }
-})
-
-// Write specification section with ALL subsections
-ba-writer.enrichSection({
-  featureId: {module_feature_id},
-  section: "specification",
-  data: {
-    actors: [{role, description, permissions}],                   // 8a
-    useCases: [{id, name, primaryActor, permission, mainScenario, linkedRules}],  // 8b
-    functionalRequirements: [{id, statement, priority, linkedRules, linkedUseCases, acceptanceCriteria}],  // 8c
-    permissionMatrix: {permissions: [], roleAssignments: []},     // 8d
-    navigation: {entries: []},                                    // 8e
-    seedDataCore: {navigationModules, navigationSections, navigationResources, navigationTranslations, permissions, rolePermissions, permissionConstants},  // 8f (7 arrays)
-    gherkinScenarios: [{feature, scenarios: [{name, tags, given, when, then}]}],  // 8g
-    validations: [{entity, field, rules, errorMessageKey}],       // 8h
-    messages: [{code, type, title, message, i18nKey}],            // 8i
-    lifeCycles: [{entity, field, initialState, states: [{id, displayName, color, allowedTransitions, isTerminal}], transitions}],  // 8j
-    apiEndpoints: [{method, path, permission, requestDto, responseDto, description}],  // 8k
-    i18nKeys: {...},                                              // 8l
-    sections: [{code, labels, route, icon, permission, wireframe, useCases, businessRules, resources}],  // from 3a-bis
-    dashboards: [{code, title, kpis, filters, refreshMode}],     // from 3d
-    uiWireframes: [{screen, section, mockup, elements, actions, componentMapping, layout}]  // from 3b
-  }
-})
-
-// Write validation results from step 9
-// (Already written in step 9e above)
-
-ba-writer.updateStatus({module_feature_id}, "specified")
-
-// Update master
-ba-writer.updateModuleStatus({feature_id}, {currentModule.code}, "specified")
-```
-
-#### 11-POST-CHECK: Verify Written Data (BLOCKING)
-
-> **CRITICAL — Data loss prevention.** After writing, READ BACK the module feature.json and verify the following arrays are **non-empty**:
-
-```javascript
-// READ BACK the written feature.json
-const written = ba-reader.read({module_feature_id});
-
-// BLOCKING checks — if ANY fails, the write was incomplete
-const checks = [
-  { key: "specification.actors",          actual: written.specification?.actors?.length,          min: 2 },
-  { key: "specification.useCases",        actual: written.specification?.useCases?.length,        min: 2 },
-  { key: "specification.wireframes",      actual: (written.specification?.uiWireframes?.length || written.specification?.wireframes?.length || 0), min: 1 },
-  { key: "specification.sections",        actual: written.specification?.sections?.length,        min: 1 },
-  { key: "specification.seedDataCore",    actual: Object.keys(written.specification?.seedDataCore || {}).length, min: 7 },
-  { key: "specification.lifeCycles",      actual: written.specification?.lifeCycles?.length,      min: 0 },
-  { key: "specification.gherkinScenarios",actual: written.specification?.gherkinScenarios?.length,min: 1 },
-  { key: "specification.apiEndpoints",    actual: written.specification?.apiEndpoints?.length,    min: 1 }
-];
-
-const failures = checks.filter(c => (c.actual || 0) < c.min);
-
-IF failures.length > 0:
-  BLOCKING ERROR: "Feature.json write INCOMPLETE — missing data in: {failures.map(f => f.key).join(', ')}"
-  → Re-execute section 11 write with ALL specification data
-  → DO NOT proceed to next module until ALL checks pass
-
-// SPECIAL CHECK: wireframes content verification (check BOTH key names)
-const wireframes = written.specification?.uiWireframes || written.specification?.wireframes || [];
-IF wireframes.length > 0:
-  const emptyMockups = wireframes.filter(wf => !wf.mockup && !wf.ascii && !wf.content);
-  IF emptyMockups.length > 0:
-    WARNING: "{emptyMockups.length} wireframes have empty mockup content — verify step-03b data"
-
-// SPECIAL CHECK: wireframes >= sections
-const sectionCount = written.specification?.sections?.length || 0;
-IF wireframes.length < sectionCount:
-  BLOCKING ERROR: "{wireframes.length} wireframes < {sectionCount} sections — wireframes MISSING for some sections"
-  → Re-read wireframes from conversation context and re-write to feature.json
-```
-
-#### 11-POST-CHECK-BASH: Acceptance Criteria Verification (BLOCKING)
-
-> **CRITICAL:** The pseudocode checks above are interpreted by the model — they can be "passed" incorrectly.
-> This bash check reads the REAL file on disk and provides an objective verification.
-> See [references/acceptance-criteria.md](../references/acceptance-criteria.md) for the full acceptance criteria definition.
-
-```bash
-MODULE_JSON="{module_feature_json_path}"
-node -e "
-const fs = require('fs');
-const data = JSON.parse(fs.readFileSync(process.argv[1], 'utf-8'));
-const spec = data.specification || {};
-const analysis = data.analysis || {};
-const wf = spec.uiWireframes || spec.wireframes || [];
-const sections = spec.sections || [];
-const checks = [
-  ['entities >= 1',          (analysis.entities||[]).length,          1],
-  ['useCases >= 2',          (spec.useCases||[]).length,              2],
-  ['FRs >= 4',               (spec.functionalRequirements||[]).length,4],
-  ['wireframes >= 1',        wf.length,                               1],
-  ['wireframes >= sections', wf.length,                               sections.length],
-  ['sections >= 1',          sections.length,                          1],
-  ['seedDataCore 7 arrays',  Object.keys(spec.seedDataCore||{}).filter(k=>(spec.seedDataCore||{})[k]&&(spec.seedDataCore||{})[k].length>0).length, 7],
-  ['gherkin is array',       Array.isArray(spec.gherkinScenarios)?1:0, 1],
-  ['apiEndpoints >= 1',      (spec.apiEndpoints||[]).length,           1],
-  ['messages >= 4',          (spec.messages||[]).length,               4],
-  ['validations >= 1',       (spec.validations||[]).length,            1],
-  ['navigation entries',     (spec.navigation?.entries||[]).length,    1],
-  ['dataLifecycle',          analysis.dataLifecycle ? 1 : 0,           1],
-  ['validation section',     data.validation ? 1 : 0,                  1]
-];
-// AC-18: i18n 4 languages check
-const i18n = spec.i18nKeys || {};
-const i18nLangs = ['fr','en','it','de'];
-let i18nMissing = 0;
-function checkI18nLeaf(obj) {
-  if (typeof obj !== 'object' || obj === null) return;
-  const hasLang = i18nLangs.some(l => typeof obj[l] === 'string');
-  if (hasLang) { i18nLangs.forEach(l => { if (!obj[l]) i18nMissing++; }); }
-  else { Object.values(obj).forEach(v => checkI18nLeaf(v)); }
-}
-checkI18nLeaf(i18n);
-checks.push(['i18n 4 languages (missing keys)', 0, i18nMissing > 10 ? 1 : 0]);
-
-const fails = checks.filter(c => c[1] < c[2]);
-fails.forEach(f => console.error('FAIL: ' + f[0] + ' = ' + f[1] + ' (min: ' + f[2] + ')'));
-if (i18nMissing > 0 && i18nMissing <= 10) { console.warn('WARNING: ' + i18nMissing + ' i18n leaf keys missing translations (IT/DE)'); }
-// Check wireframe content
-const emptyWf = wf.filter(w => !w.mockup && !w.ascii && !w.content);
-if (emptyWf.length > 0) { fails.push(['wireframe content', 0, 1]); console.error('FAIL: ' + emptyWf.length + ' wireframes have EMPTY content'); }
-// Check entity attribute types
-const badAttrs = (analysis.entities||[]).flatMap(e => (e.attributes||[]).filter(a => !a.type).map(a => e.name+'.'+a.name));
-if (badAttrs.length > 0) { fails.push(['attr.type', 0, 1]); console.error('FAIL: attributes without type: ' + badAttrs.join(', ')); }
-// AC-15: Validation rules format
-const badRules = (spec.validations||[]).filter(v => v.rules && !Array.isArray(v.rules));
-if (badRules.length > 0) { fails.push(['AC-15: rules not array', badRules.length, 0]); console.error('FAIL: AC-15: ' + badRules.length + ' validations have rules as string'); }
-// AC-16: Messages must have message field
-const noMsg = (spec.messages||[]).filter(m => !m.message);
-if (noMsg.length > 0) { fails.push(['AC-16: message missing', noMsg.length, 0]); console.error('FAIL: AC-16: ' + noMsg.length + ' messages missing message field'); }
-// AC-17: Gherkin content structure
-if (Array.isArray(spec.gherkinScenarios)) { const badG = spec.gherkinScenarios.filter(g => !g.feature || !Array.isArray(g.scenarios)); if (badG.length > 0) { fails.push(['AC-17: gherkin content', badG.length, 0]); console.error('FAIL: AC-17: ' + badG.length + ' gherkin entries invalid'); } }
-if (fails.length > 0) { console.error('BLOCKING: ' + fails.length + ' acceptance criteria failed'); process.exit(1); }
-console.log('PASS: All acceptance criteria met');
-" "$MODULE_JSON"
-```
-
-IF this check FAILS:
-- Identify which criteria failed from the output
-- Fix the corresponding data (re-read from conversation context, re-generate, or re-run the step that produces it)
-- Re-write to feature.json
-- Re-run the POST-CHECK until PASS
-
-> **WHY:** Step-03b now writes wireframes intermediately (since fix), but step-03d section 11 does the FULL write.
-> This bash POST-CHECK is the ultimate safety net — it reads the REAL file and verifies ALL acceptance criteria.
-> The pseudocode checks (section 9a) catch issues early; this bash check catches anything the model missed.
-
----
-
-### 11-bis. Deploy Incremental Interactive HTML (MANDATORY)
-
-> **After each module is specified, deploy/update the interactive HTML document with all available data.**
-> The client can review completed modules while the next module is being specified.
-
-See [references/validate-incremental-html.md](../references/validate-incremental-html.md) for the complete deployment process (read sources, build FEATURE_DATA, replace placeholders, write HTML).
-
-Uses the **same mapping** as step-05b-deploy.md — only difference is `moduleSpecs` only includes completed modules.
-
----
-
-### 12. Loop Decision
-
-> **TEAM AGENT MODE: SKIP THIS ENTIRE SECTION.** If you are a team agent (your prompt contains `PROPOSE & REVIEW`), go directly to **section 12-bis** below. Section 12 is for inline mode ONLY. You MUST NOT load the next step or advance the module loop — the team lead handles orchestration.
-
-**INLINE MODE ONLY (main conversation):**
-
-```
-ba-writer.advanceModuleLoop({feature_id})
-→ Increments currentModuleIndex
-→ Updates completedModules
-→ Updates currentModule
-
-IF currentModuleIndex < moduleOrder.length:
-  Display: "→ Module suivant: {nextModule} ({currentModuleIndex + 1}/{total})"
-  Load: steps/step-03a1-setup.md
-
-IF currentModuleIndex >= moduleOrder.length:
-  ba-writer.updateStatus({feature_id}, "specified")
-
-  // CHECKPOINT: Save progress BEFORE transition (protects against context exhaustion)
-  ba-writer.enrichSection({
-    featureId: {feature_id},
-    section: "metadata.workflow",
-    data: {
-      lastCompletedStep: "step-03d-validate",
-      allModulesSpecified: true
-    }
-  })
-
-  // CHECK: Are we in project mode (multi-application)?
-  IF workflow.mode === "project":
-    // APPLICATION LOOP: Check if more applications remain
-    projectFeature = ba-reader.findProjectFeature()
-    ba-writer.updateApplicationStatus(projectId, currentApp.code, "specified")
-    ba-writer.advanceApplicationLoop(projectId)
-
-    IF projectFeature.metadata.workflow.currentApplicationIndex + 1 < projectFeature.metadata.workflow.applicationOrder.length:
-      nextApp = projectFeature.metadata.workflow.applicationOrder[currentApplicationIndex + 1]
-      Display: "═══ Application {currentApp.name} terminée ({N}/{total}). Prochaine : {nextApp} ═══"
-      Display: "⚠ NE PAS lancer /ralph-loop — d'autres applications et la consolidation sont encore nécessaires."
-      // Load step-02 for next application's module decomposition
-      Load: steps/step-02-decomposition.md
-    ELSE:
-      Display: "═══ Toutes les applications spécifiées! Passage à la consolidation cross-application... ═══"
-      Display: "⚠ NE PAS lancer /ralph-loop — la consolidation et le handoff sont encore nécessaires."
-      ba-writer.updateStatus(projectId, "specified")
-      Load: steps/step-04a-collect.md
-  ELSE:
-    // SINGLE APPLICATION: Current behavior
-    Display: "═══ Tous les modules spécifiés! Passage à la consolidation (04a→04c) puis handoff (05a→05c)... ═══"
-    Display: "⚠ NE PAS lancer /ralph-loop — la consolidation et le handoff sont encore nécessaires."
-
-    // MANDATORY TRANSITION — DO NOT STOP HERE
-    // If context is near exhaustion, the checkpoint above ensures the user
-    // can resume with /business-analyse and it will route to step-04a automatically.
-    Load: steps/step-04a-collect.md
-```
-
----
-
-## 12-bis. Agent Mode: Send Proposal to Team Lead
-
-**IF running as TEAM AGENT (autonomous mode — detected in step-03a1 MODE DETECTION):**
-
-After all validation checks PASS and module feature.json is written:
-
-### A. Build Structured Summary
-
-Read the written module feature.json via ba-reader and build:
-
-```
-## Module: {moduleCode} — Specification Summary
-
-### Entities ({count})
-| Entity | Key Attributes | Relationships |
-|--------|---------------|---------------|
-| {name} | {attr1} ({type}), {attr2} ({type}), ... | → {relatedEntity} (FK) |
-
-### Sections ({count})
-| Section | Description | Resources |
-|---------|-------------|-----------|
-| {code} | {description} | {resource1}, {resource2}, ... |
-
-### Use Cases ({count})
-- {UC-ID}: {title}
-
-### Business Rules ({count})
-- {BR-ID}: {title} ({category})
-
-### Permissions
-- {permission.path}: {roles with access}
-
-### Seed Data Counts
-- Modules: {n} | Sections: {n} | Resources: {n}
-- Translations: {n} | Permissions: {n} | Role mappings: {n} | Constants: {n}
-
-### Wireframes
-- {section}: {1-line description of layout and key components}
-```
-
-### B. Send Proposal
-
-```
-SendMessage({
-  type: "message",
-  recipient: "team-lead",
-  content: "PROPOSAL_READY:{moduleCode}\n\n{structured_summary}",
-  summary: "{moduleCode} specification ready for review"
-})
-```
-
-### C. Wait for Response
-
-**WAIT** for team lead to respond:
-
-- **`APPROVED:{moduleCode}`** → Proceed to step D (Module Complete)
-- **`REVISION:{moduleCode}\n{feedback}`** → Apply changes:
-  1. Read the feedback
-  2. Update the relevant sections in feature.json via ba-writer
-  3. Re-run validation checks (section 9)
-  4. Re-build summary and send new `PROPOSAL_READY`
-  5. Max **3 revision cycles** — after 3, send `ERROR:{moduleCode}:Max revisions reached` and stop
-
-### D. Module Complete
-
-After approval:
-
-```
-ba-writer.updateModuleStatus({feature_id}, {moduleCode}, "specified")
-```
-
-```
-SendMessage({
-  type: "message",
-  recipient: "team-lead",
-  content: "MODULE_COMPLETE:{moduleCode}",
-  summary: "{moduleCode} module complete"
-})
-```
-
-Then WAIT — the team lead will send you a `shutdown_request`.
-Do NOT loop to next module. The team lead handles module ordering and will spawn a new agent for the next module.
-
-### E. Shutdown (MANDATORY)
-
-When you receive a `shutdown_request` from the team lead:
-
-```
-SendMessage({
-  type: "shutdown_response",
-  request_id: "{requestId from the shutdown_request message}",
-  approve: true
-})
-```
-
-This terminates your process. Do NOT output any text after `shutdown_response`.
-
-**Sequence reminder:** APPROVED → MODULE_COMPLETE → shutdown_request → shutdown_response (4 separate steps, never skip).
-
----
-
-## SELF-VERIFICATION (MANDATORY before loading next step)
-
-Before proceeding to step-03a (next module) or step-04 (consolidation), VERIFY:
-
-1. **Module feature.json has specification section** with ALL sub-sections (actors, useCases, functionalRequirements, permissionMatrix, navigation, seedDataCore, gherkinScenarios, validations, messages, lifeCycles, apiEndpoints, i18nKeys, sections, uiWireframes)
-2. **Module feature.json status** = "specified" (set by section 11)
-3. **Master modules[].status** for this module = "specified" (set by section 11)
-4. **ba-interactive.html exists** and includes data for this module (set by section 11-bis)
-
-**IF any check fails → FIX before proceeding.** Do NOT advance the module loop with incomplete data.
-
----
-
-## SINGLE-MODULE MODE
-
-When only 1 module in moduleOrder:
-
-- Same full validation flow
-- Skip cross-module references (section 5)
-- After validation → direct to step-04a-collect.md (auto-consolidation)
-
----
-
-## RESUME SUPPORT
-
-IF the step is resumed (e.g., after interruption):
-
-```
-Read metadata.workflow.currentModuleIndex
-Read metadata.workflow.currentModule
-→ Resume at the correct module
-→ If module has partial data, show what exists and continue from there
-```
-
----
-
-## CONTEXT MANAGEMENT
-
-| Loaded | Not Loaded |
-|--------|------------|
-| _shared.md | Other module feature.json (full) |
-| feature-schema.json | Questionnaires not needed |
-| Questionnaires for this module | Templates |
-| Completed modules summary (100 lines max) | React viewer files |
-
----
-
-## STATE VARIABLES
-
-| Variable | Description |
-|----------|-------------|
-| `{currentModule}` | Current module being specified |
-| `{currentModuleIndex}` | Index in moduleOrder |
-| `{completedModules}` | Array of already specified module codes |
-| `{module_feature_id}` | Feature ID for the module-level feature.json |
-
----
-
-## SUCCESS METRICS
-
-- Module specification validation PASS
-- Per-module validation results recorded in feature.json
-- Client confirmed module specification
-- Module feature.json written with all sections
-- Master updated with module status
-- Interactive HTML deployed/updated with completed module data (incremental)
-- Loop advanced correctly or consolidation phase initiated
-
----
-
-## FAILURE MODES
-
-- Validation FAIL → offer auto-correct / manual revise / force
-- Client rejects → revise specific sections via step-03a or step-03b
-- Cross-module reference error → clarify with client
-- Interruption → resumable via workflow state
-
----
-
-## NEXT STEP
-
-Conditional:
-
-- IF modules remaining → Load `steps/step-03a1-setup.md`
-- IF all modules done → Load `steps/step-04a-collect.md`
+---
+name: step-03d-validate
+description: Per-module validation, write feature.json, incremental HTML deploy, loop decision
+model: sonnet
+next_step: steps/step-03a1-setup.md OR steps/step-04a-collect.md (conditional)
+---
+
+> **Context files:** `_shared.md` | `_module-loop.md`
+
+# Step 3d: Validation & Loop
+
+## MANDATORY EXECUTION RULES
+
+- ALWAYS use ULTRATHINK mode
+- This step VALIDATES the specification from step-03c, writes it to feature.json, and decides loop continuation
+- ALWAYS verify specification completeness before writing
+- ALL communication in `{language}`
+
+## MODE DETECTION (inherited from step-03a1)
+
+> **CRITICAL: Re-check your execution mode before proceeding.**
+
+**IF you are running as a TEAM AGENT** (your prompt contains `PROPOSE & REVIEW` or `team-lead` as recipient):
+  → **NEVER** use `AskUserQuestion` in ANY section below (sections 9d, 10, 12)
+  → Section 9d: if validation fails, AUTO-CORRECT silently (no user options)
+  → Section 12: **SKIP ENTIRELY** — go directly to section 12-bis (Agent Mode)
+  → After writing feature.json, send `PROPOSAL_READY` to team lead via SendMessage
+  → NEVER present options/menus to the user — you are an autonomous agent
+
+**IF you are running in the MAIN CONVERSATION** (classic inline mode):
+  → Normal interactive mode — use `AskUserQuestion` as documented
+  → Section 12-bis does NOT apply to you
+- **ID NAMING RULE (MANDATORY, NO EXCEPTION):**
+  All IDs MUST include a module prefix to guarantee application-wide uniqueness.
+  The prefix is derived from the module code initials (2-4 chars):
+    UserManagement → UM | VehicleManagement → VM | PartsInventory → PI
+    RepairManagement → RM | MaintenanceSchedule → MS | DataSync → DS
+    Notifications → NT | Dashboard → DB | Orders → OR | Customers → CU
+
+  NEVER use bare IDs (UC-001, BR-VAL-001) in multi-module mode.
+
+## YOUR TASK
+
+Validate the module specification for completeness and consistency, write to feature.json, deploy incremental HTML, and decide whether to loop to the next module or proceed to consolidation.
+
+---
+
+### 9. Per-Module Validation
+
+#### 9a. Completeness Checks
+
+> **CRITICAL:** Checks MUST count ACTUAL elements in arrays, NOT declarative values.
+> A check that reports PASS when the array is empty is a LIE and BLOCKS downstream quality.
+
+| Section | Minimum | How to verify | Status |
+|---------|---------|---------------|--------|
+| actors | 2 | `specification.actors.length >= 2` | PASS/FAIL |
+| useCases | 2 | `specification.useCases.length >= 2` | PASS/FAIL |
+| functionalRequirements | 4 | `specification.functionalRequirements.length >= 4` | PASS/FAIL |
+| permissionMatrix | 1 resource × 2 roles | `specification.permissionMatrix.permissions.length >= 1 && specification.permissionMatrix.roleAssignments.length >= 2` | PASS/FAIL |
+| entities | 1 | `analysis.entities.length >= 1` | PASS/FAIL |
+| entitySchemaFormat | attributes[] not fields[] (BLOCKING) | `analysis.entities.every(e => e.attributes?.length > 0)` | PASS/FAIL |
+| entityAttributeTypes | ALL attributes have `type` field (BLOCKING) | `analysis.entities.every(e => e.attributes.every(a => a.type))` — **NOTE:** Entities in analysis phase may NOT have `type`. Step-03c auto-fix MUST have run first. If types are missing, RE-RUN step-03c auto-fix algorithm. | PASS/FAIL |
+| navigation | entries[] present (BLOCKING) | `specification.navigation?.entries?.length >= 1` | PASS/FAIL |
+| dataLifecycle | present in analysis (BLOCKING) | `analysis.dataLifecycle !== undefined` | PASS/FAIL |
+| wireframes | 1 per section (BLOCKING) | `(specification.uiWireframes \|\| specification.wireframes \|\| []).length >= (specification.sections \|\| []).length` (count REAL elements, check BOTH key names) | PASS/FAIL |
+| wireframeSchema | All required fields present (BLOCKING) | `(specification.uiWireframes \|\| specification.wireframes \|\| []).every(w => (w.screen \|\| w.title) && w.section && (w.mockup \|\| w.ascii \|\| w.content))` | PASS/FAIL |
+| sections | 1 (BLOCKING) | `specification.sections.length >= 1` (EVERY module needs at least 1 section) | PASS/FAIL |
+| gherkinScenarios | 1 array entry | `Array.isArray(specification.gherkinScenarios) && specification.gherkinScenarios.length >= 1` | PASS/FAIL |
+| gherkinFormat | Array not object (BLOCKING) | `Array.isArray(specification.gherkinScenarios)` (NOT a single object) | PASS/FAIL |
+| validations | 1 | `specification.validations.length >= 1` | PASS/FAIL |
+| validationFormat | rules[] array (BLOCKING) | `specification.validations.every(v => Array.isArray(v.rules))` (NOT singular `rule`) | PASS/FAIL |
+| messages | 4 | `specification.messages.length >= 4` | PASS/FAIL |
+| messageFormat | `message` field present (BLOCKING) | `specification.messages.every(m => m.message)` | PASS/FAIL |
+| lifeCycles | 1 (if entity has status) | `specification.lifeCycles.length >= 1` (if any entity has status/state field) | PASS/FAIL |
+| seedDataCore | 9 arrays present with content | See detailed check below | PASS/FAIL (BLOCKING) |
+| apiEndpoints | 1 | `specification.apiEndpoints.length >= 1` | PASS/FAIL |
+| i18nKeys | present | `specification.i18nKeys !== undefined && specification.i18nKeys !== null` | PASS/FAIL |
+| navigationIcons | non-null | `specification.seedDataCore.navigationModules.every(m => m.icon !== null)` | PASS/FAIL |
+
+**seedDataCore detailed check (BLOCKING):**
+```javascript
+const sdc = specification.seedDataCore;
+const currentModuleIndex = metadata.workflow?.currentModuleIndex || 0;
+const checks = [
+  // navigationApplications and applicationRoles: required for first module (index 0), can be empty for subsequent modules
+  { key: "navigationApplications", actual: sdc.navigationApplications?.length || 0, min: currentModuleIndex === 0 ? 1 : 0 },
+  { key: "applicationRoles",      actual: sdc.applicationRoles?.length || 0,       min: currentModuleIndex === 0 ? 1 : 0 },
+  { key: "navigationModules",      actual: sdc.navigationModules?.length || 0,      min: 1 },
+  { key: "navigationSections",     actual: sdc.navigationSections?.length || 0,     min: 1 },  // EVERY module needs ≥1 section
+  { key: "navigationResources",    actual: sdc.navigationResources?.length || 0,    min: 1 },
+  { key: "navigationTranslations", actual: sdc.navigationTranslations?.length || 0, min: 2 },  // min fr+en
+  { key: "permissions",            actual: sdc.permissions?.length || 0,            min: 1 },
+  { key: "rolePermissions",        actual: sdc.rolePermissions?.length || 0,        min: 1 },
+  { key: "permissionConstants",    actual: sdc.permissionConstants?.length || 0,    min: 1 }
+];
+const failures = checks.filter(c => c.actual < c.min);
+IF failures.length > 0:
+  BLOCKING ERROR: "seedDataCore incomplete — empty arrays: {failures.map(f => f.key).join(', ')}"
+  → Fix: Ensure specification.sections[] has ≥1 entry, then re-run 8f-bis transform
+```
+
+#### 9b. Consistency Checks
+
+- Every UC has ≥1 linked FR
+- Every FR has ≥1 linked BR
+- All BR references exist in analysis.businessRules
+- All actors appear in permissionMatrix
+- Permission paths use correct format: module-level `{app}.{module}.{action}` or section-level `{app}.{module}.{section}.{action}`
+- rolePermissions paths match permissions paths
+- API routes use consistent prefix
+
+#### 9c. Convention Checks
+
+- UC-NNN format
+- FR-NNN format
+- BR-{CATEGORY}-NNN format
+- Entity names PascalCase
+- Field names camelCase
+- Entity attribute format: `attributes[]` with {name, description}, NOT `fields[]` with {name, type} — entities must NOT have tableName or primaryKey
+- Wireframe structure: `screen` (not `name`/`title`), `mockup` (not `ascii`/`content`), `componentMapping` is array of {wireframeElement, reactComponent} (not plain key-value object), `layout` is object with regions (not string)
+- Wireframe field naming: uses `screen` (not `title`), `mockup` (not `ascii`) — auto-fix if wrong (see 9c-fix below)
+- Permission paths dot-separated lowercase
+
+#### 9c-fix. Auto-Fix Wireframe Field Names (MANDATORY before writing)
+
+> **CRITICAL:** The agent may use non-canonical field names (`title`, `ascii`, `name`, `content`).
+> These MUST be normalized to canonical names before writing to feature.json.
+
+```javascript
+// AUTO-FIX: Normalize wireframe field names before writing
+const wireframes = specification.uiWireframes || specification.wireframes || [];
+for (const wf of wireframes) {
+  if (wf.title && !wf.screen) { wf.screen = wf.title; delete wf.title; }
+  if (wf.ascii && !wf.mockup) { wf.mockup = wf.ascii; delete wf.ascii; }
+  if (wf.content && !wf.mockup) { wf.mockup = wf.content; delete wf.content; }
+  if (wf.name && !wf.screen) { wf.screen = wf.name; delete wf.name; }
+}
+// Store normalized wireframes back under canonical key
+specification.uiWireframes = wireframes;
+```
+
+#### 9c-fix-entities. Auto-Fix Entity Attribute Types (MANDATORY before writing)
+
+> **DEFENSE-IN-DEPTH:** step-03c POST-CHECK should catch attributes without `type`,
+> but if any slip through to step-03d, auto-fix them here before writing to feature.json.
+> Uses the same type inference algorithm as step-03c POST-CHECK.
+
+```javascript
+// AUTO-FIX: Ensure all entity attributes have structured type
+const entities = analysis.entities || [];
+let entityAutoFixCount = 0;
+for (const entity of entities) {
+  for (const attr of entity.attributes || []) {
+    if (!attr.type) {
+      // Infer type from attribute name pattern
+      if (attr.validation?.match(/max\s*\d+/i) || attr.name.match(/name|title|code|description|label|email|phone|address/i)) {
+        attr.type = "string";
+        const maxMatch = attr.validation?.match(/max\s*(\d+)/i);
+        if (maxMatch) attr.maxLength = parseInt(maxMatch[1]);
+      } else if (attr.name.match(/id$/i)) {
+        attr.type = "Guid";
+      } else if (attr.name.match(/date|At$/i)) {
+        attr.type = "DateTime";
+      } else if (attr.name.match(/is[A-Z]|has[A-Z]|active|enabled/)) {
+        attr.type = "bool";
+      } else if (attr.name.match(/amount|salary|rate|price|total/i)) {
+        attr.type = "decimal";
+      } else if (attr.name.match(/count|number|order|sort|index/i)) {
+        attr.type = "int";
+      } else {
+        attr.type = "string"; // safe default
+      }
+      // Normalize free-text maxLength
+      if (typeof attr.validation === 'string' && !attr.maxLength) {
+        const m = attr.validation.match(/max\s*(\d+)/i);
+        if (m) attr.maxLength = parseInt(m[1]);
+      }
+      if (attr.required === undefined) attr.required = true;
+      entityAutoFixCount++;
+    }
+  }
+}
+if (entityAutoFixCount > 0) {
+  console.warn(`DEFENSE-IN-DEPTH: auto-fixed ${entityAutoFixCount} attributes without type in step-03d`);
+}
+```
+
+#### 9d. Decision
+
+> **TEAM AGENT MODE:** If running as team agent, SKIP AskUserQuestion. On FAIL → auto-correct silently. On PASS → proceed directly to section 11 (write). NEVER present options to the user.
+
+**INLINE MODE ONLY (main conversation):**
+
+IF validation PASS:
+  Display summary, ask client for confirmation
+
+IF validation FAIL:
+  Display errors, offer:
+  ```
+  options:
+    - label: "Auto-corriger"
+      description: "Corriger automatiquement les erreurs détectées"
+    - label: "Réviser manuellement"
+      description: "Revenir aux sections problématiques"
+    - label: "Forcer la validation"
+      description: "Accepter malgré les erreurs (risqué)"
+  ```
+
+#### 9e. Write Module Validation Section (MANDATORY)
+
+> **The module feature.json MUST have a `validation` section recording the results of steps 9a-9d.**
+
+```
+ba-writer.enrichSection({
+  featureId: {module_feature_id},
+  section: "validation",
+  data: {
+    validatedAt: "{ISO timestamp}",
+    completenessChecks: [
+      { "section": "useCases", "count": {count}, "minimum": 2, "status": "PASS|FAIL" },
+      { "section": "functionalRequirements", "count": {count}, "minimum": 4, "status": "PASS|FAIL" },
+      { "section": "wireframes", "count": {count}, "minimum": {section_count}, "status": "PASS|FAIL" },
+      { "section": "seedDataCore", "count": 7, "minimum": 7, "status": "PASS|FAIL" }
+    ],
+    consistencyChecks: [
+      { "check": "UC-FR linkage", "passed": {n}, "warnings": {n}, "errors": {n}, "status": "PASS|FAIL" },
+      { "check": "BR references valid", "passed": {n}, "warnings": {n}, "errors": {n}, "status": "PASS|FAIL" },
+      { "check": "Permission path format", "passed": {n}, "warnings": {n}, "errors": {n}, "status": "PASS|FAIL" }
+    ],
+    conventionChecks: [
+      { "check": "ID naming with module prefix", "status": "PASS|FAIL", "details": "..." },
+      { "check": "Entity names PascalCase", "status": "PASS|FAIL", "details": "..." }
+    ],
+    semanticChecks: [
+      { "check": "orphan-permissions", "module": "{currentModule}", "status": "PASS|WARNING|ERROR", "details": "...", "autoFixed": false },
+      { "check": "uc-sans-fr", "module": "{currentModule}", "status": "PASS|WARNING|ERROR", "details": "...", "autoFixed": false },
+      { "check": "missing-wireframes", "module": "{currentModule}", "status": "PASS|WARNING|ERROR", "details": "...", "autoFixed": false }
+    ],
+    warnings: ["List of non-blocking warnings"],
+    decision: {
+      "approved": true,
+      "reason": "All checks passed",
+      "approvalMode": "standard",
+      "approvedBy": "Client",
+      "approvedAt": "{ISO timestamp}"
+    }
+  }
+})
+```
+
+#### 9e-POST-CHECK: Validation Section Persistence (BLOCKING)
+
+> **CRITICAL:** The `validation` section MUST be persisted to feature.json. Without it, consolidation checks fail.
+
+```bash
+MODULE_JSON="{module_feature_json_path}"
+node -e "const d=JSON.parse(require('fs').readFileSync(process.argv[1],'utf-8'));
+if(!d.validation||!d.validation.decision) { console.error('FAIL: validation section missing or incomplete'); process.exit(1); }
+console.log('PASS: validation section present with decision');" "$MODULE_JSON"
+```
+
+IF this check FAILS → re-execute section 9e write and re-run this check.
+
+#### 9f. Module Specification Checklist (BLOCKING)
+
+> **CRITICAL:** This checklist MUST be FULLY COMPLETED before marking module status = "specified".
+
+**Execute the comprehensive validation checklist:**
+
+Run the 29-check validation process across 10 categories:
+- Data Model (4 checks) | Business Rules (3 checks) | Use Cases & FRs (4 checks)
+- Permissions (3 checks) | UI & Navigation (4 checks) | I18N & Messages (3 checks)
+- Seed Data (2 checks) | API Endpoints (2 checks) | Validations (1 check) | Gherkin (3 checks)
+
+```javascript
+const blockingFailures = checks.filter(c => c.blocking && c.status === "FAIL");
+
+IF blockingFailures.length > 0:
+  // CONDITIONAL LOAD: Only load detailed checklist on validation failure
+  Read references/validation-checklist.md
+  Display:
+    - Full validation logic for FAILED checks
+    - Exact requirements (minimums, patterns, formats)
+    - Fix recommendations for each failure
+
+  BLOCKING ERROR: Module specification incomplete
+  → Fix all failures before proceeding
+
+ELSE:
+  All blocking checks passed ✓
+  Display: "{warnings_count} warnings (non-blocking)"
+  → Proceed to mark module as "specified"
+```
+
+**Optimization:** The detailed 280-line validation-checklist.md is loaded **only when validation fails** (saves ~10,000 tokens on success path).
+
+---
+
+### 9g. Anti-Premature-Completion Guard (MANDATORY)
+
+> **CRITICAL — NEVER say "the analysis is complete" or "ready for /ralph-loop" at this point.**
+> Step-03d is the END of SPECIFICATION, NOT the end of the BA workflow.
+> Steps 04a → 04b → 04c (consolidation) + 05a → 05b → 05c (handoff) are STILL REQUIRED.
+> Without handoff, /ralph-loop will generate an INCOMPLETE PRD (no frontend, no tests).
+
+**FORBIDDEN phrases in step-03d output:**
+- "L'analyse métier est complète" / "The business analysis is complete"
+- "Prêt pour /ralph-loop" / "Ready for /ralph-loop"
+- "Vous pouvez maintenant lancer /ralph-loop"
+- Any variation suggesting the BA is finished or /ralph-loop can be invoked
+
+**REQUIRED phrasing:**
+- "Module {name} spécifié. {Remaining} modules restants." (if more modules)
+- "Tous les modules spécifiés. Passage à la consolidation (étapes 04a-04c)..." (if last module)
+- NEVER mention /ralph-loop — only step-05c (ralph readiness) validates the handoff
+
+### 10. Module Summary with Roles & Permissions
+
+Display comprehensive summary:
+
+```
+═══════════════════════════════════════════════════════════
+  MODULE SPÉCIFIÉ: {currentModule} ({currentModuleIndex + 1}/{total})
+═══════════════════════════════════════════════════════════
+
+| Aspect | Valeur |
+|--------|--------|
+| Entités | {count} ({entity names}) |
+| Business Rules | {count} (VAL: X, CALC: Y, WF: Z) |
+| Use Cases | {count} (UC-001 → UC-NNN) |
+| Exigences | {count} FRs |
+| Sections | {count} ({section names}) |
+| Maquettes | {count} wireframes |
+| Permissions | {count} paths |
+| Tests Gherkin | {count} scénarios |
+| Messages | {count} |
+
+### Rôles & Permissions pour {currentModule}
+| Rôle | Permissions |
+|------|------------|
+| {App} Admin | {app}.{module}.* |
+| {App} Manager | read, create, update, assign |
+| {App} Contributor | read, create, update |
+| {App} Viewer | read |
+
+→ Validation: {PASS/FAIL}
+═══════════════════════════════════════════════════════════
+
+{IF validation PASS}:
+  → Module validé automatiquement
+  → Passage automatique au module suivant (section 12)
+
+{IF validation FAIL}:
+  → Afficher les erreurs et offrir les options de correction (section 9d)
+```
+
+---
+
+### 11. Write Module Feature.json
+
+```
+// Write analysis section with ALL subsections
+ba-writer.enrichSection({
+  featureId: {module_feature_id},
+  section: "analysis",
+  data: {
+    objectives: [{id, objective, metric, target}],       // from 6a
+    entities: [{name, description, attributes, relationships}],  // from 6b — SCHEMA FORMAT
+    businessRules: [{id, name, category, statement, priority, conditions, examples, testability}],  // from 7 — SCHEMA FORMAT
+    processFlow: {entryPoints, mainFlow, decisionPoints, alternativeFlows},  // from 6c
+    dataLifecycle: {retentionPeriod, archiveStrategy, gdprCompliance, states}  // from 6d
+  }
+})
+
+// Write specification section with ALL subsections
+ba-writer.enrichSection({
+  featureId: {module_feature_id},
+  section: "specification",
+  data: {
+    actors: [{role, description, permissions}],                   // 8a
+    useCases: [{id, name, primaryActor, permission, mainScenario, linkedRules}],  // 8b
+    functionalRequirements: [{id, statement, priority, linkedRules, linkedUseCases, acceptanceCriteria}],  // 8c
+    permissionMatrix: {permissions: [], roleAssignments: []},     // 8d
+    navigation: {entries: []},                                    // 8e
+    seedDataCore: {navigationModules, navigationSections, navigationResources, navigationTranslations, permissions, rolePermissions, permissionConstants},  // 8f (7 arrays)
+    gherkinScenarios: [{feature, scenarios: [{name, tags, given, when, then}]}],  // 8g
+    validations: [{entity, field, rules, errorMessageKey}],       // 8h
+    messages: [{code, type, title, message, i18nKey}],            // 8i
+    lifeCycles: [{entity, field, initialState, states: [{id, displayName, color, allowedTransitions, isTerminal}], transitions}],  // 8j
+    apiEndpoints: [{method, path, permission, requestDto, responseDto, description}],  // 8k
+    i18nKeys: {...},                                              // 8l
+    sections: [{code, labels, route, icon, permission, wireframe, useCases, businessRules, resources}],  // from 3a-bis
+    dashboards: [{code, title, kpis, filters, refreshMode}],     // from 3d
+    uiWireframes: [{screen, section, mockup, elements, actions, componentMapping, layout}]  // from 3b
+  }
+})
+
+// Write validation results from step 9
+// (Already written in step 9e above)
+
+ba-writer.updateStatus({module_feature_id}, "specified")
+
+// Update master
+ba-writer.updateModuleStatus({feature_id}, {currentModule.code}, "specified")
+```
+
+#### 11-POST-CHECK: Verify Written Data (BLOCKING)
+
+> **CRITICAL — Data loss prevention.** After writing, READ BACK the module feature.json and verify the following arrays are **non-empty**:
+
+```javascript
+// READ BACK the written feature.json
+const written = ba-reader.read({module_feature_id});
+
+// BLOCKING checks — if ANY fails, the write was incomplete
+const checks = [
+  { key: "specification.actors",          actual: written.specification?.actors?.length,          min: 2 },
+  { key: "specification.useCases",        actual: written.specification?.useCases?.length,        min: 2 },
+  { key: "specification.wireframes",      actual: (written.specification?.uiWireframes?.length || written.specification?.wireframes?.length || 0), min: 1 },
+  { key: "specification.sections",        actual: written.specification?.sections?.length,        min: 1 },
+  { key: "specification.seedDataCore",    actual: Object.keys(written.specification?.seedDataCore || {}).length, min: 7 },
+  { key: "specification.lifeCycles",      actual: written.specification?.lifeCycles?.length,      min: 0 },
+  { key: "specification.gherkinScenarios",actual: written.specification?.gherkinScenarios?.length,min: 1 },
+  { key: "specification.apiEndpoints",    actual: written.specification?.apiEndpoints?.length,    min: 1 }
+];
+
+const failures = checks.filter(c => (c.actual || 0) < c.min);
+
+IF failures.length > 0:
+  BLOCKING ERROR: "Feature.json write INCOMPLETE — missing data in: {failures.map(f => f.key).join(', ')}"
+  → Re-execute section 11 write with ALL specification data
+  → DO NOT proceed to next module until ALL checks pass
+
+// SPECIAL CHECK: wireframes content verification (check BOTH key names)
+const wireframes = written.specification?.uiWireframes || written.specification?.wireframes || [];
+IF wireframes.length > 0:
+  const emptyMockups = wireframes.filter(wf => !wf.mockup && !wf.ascii && !wf.content);
+  IF emptyMockups.length > 0:
+    WARNING: "{emptyMockups.length} wireframes have empty mockup content — verify step-03b data"
+
+// SPECIAL CHECK: wireframes >= sections
+const sectionCount = written.specification?.sections?.length || 0;
+IF wireframes.length < sectionCount:
+  BLOCKING ERROR: "{wireframes.length} wireframes < {sectionCount} sections — wireframes MISSING for some sections"
+  → Re-read wireframes from conversation context and re-write to feature.json
+```
+
+#### 11-POST-CHECK-BASH: Acceptance Criteria Verification (BLOCKING)
+
+> **CRITICAL:** The pseudocode checks above are interpreted by the model — they can be "passed" incorrectly.
+> This bash check reads the REAL file on disk and provides an objective verification.
+> See [references/acceptance-criteria.md](../references/acceptance-criteria.md) for the full acceptance criteria definition.
+
+```bash
+MODULE_JSON="{module_feature_json_path}"
+node -e "
+const fs = require('fs');
+const data = JSON.parse(fs.readFileSync(process.argv[1], 'utf-8'));
+const spec = data.specification || {};
+const analysis = data.analysis || {};
+const wf = spec.uiWireframes || spec.wireframes || [];
+const sections = spec.sections || [];
+const checks = [
+  ['entities >= 1',          (analysis.entities||[]).length,          1],
+  ['useCases >= 2',          (spec.useCases||[]).length,              2],
+  ['FRs >= 4',               (spec.functionalRequirements||[]).length,4],
+  ['wireframes >= 1',        wf.length,                               1],
+  ['wireframes >= sections', wf.length,                               sections.length],
+  ['sections >= 1',          sections.length,                          1],
+  ['seedDataCore 7 arrays',  Object.keys(spec.seedDataCore||{}).filter(k=>(spec.seedDataCore||{})[k]&&(spec.seedDataCore||{})[k].length>0).length, 7],
+  ['gherkin is array',       Array.isArray(spec.gherkinScenarios)?1:0, 1],
+  ['apiEndpoints >= 1',      (spec.apiEndpoints||[]).length,           1],
+  ['messages >= 4',          (spec.messages||[]).length,               4],
+  ['validations >= 1',       (spec.validations||[]).length,            1],
+  ['navigation entries',     (spec.navigation?.entries||[]).length,    1],
+  ['dataLifecycle',          analysis.dataLifecycle ? 1 : 0,           1],
+  ['validation section',     data.validation ? 1 : 0,                  1]
+];
+// AC-18: i18n 4 languages check
+const i18n = spec.i18nKeys || {};
+const i18nLangs = ['fr','en','it','de'];
+let i18nMissing = 0;
+function checkI18nLeaf(obj) {
+  if (typeof obj !== 'object' || obj === null) return;
+  const hasLang = i18nLangs.some(l => typeof obj[l] === 'string');
+  if (hasLang) { i18nLangs.forEach(l => { if (!obj[l]) i18nMissing++; }); }
+  else { Object.values(obj).forEach(v => checkI18nLeaf(v)); }
+}
+checkI18nLeaf(i18n);
+checks.push(['i18n 4 languages (missing keys)', 0, i18nMissing > 10 ? 1 : 0]);
+
+const fails = checks.filter(c => c[1] < c[2]);
+fails.forEach(f => console.error('FAIL: ' + f[0] + ' = ' + f[1] + ' (min: ' + f[2] + ')'));
+if (i18nMissing > 0 && i18nMissing <= 10) { console.warn('WARNING: ' + i18nMissing + ' i18n leaf keys missing translations (IT/DE)'); }
+// Check wireframe content
+const emptyWf = wf.filter(w => !w.mockup && !w.ascii && !w.content);
+if (emptyWf.length > 0) { fails.push(['wireframe content', 0, 1]); console.error('FAIL: ' + emptyWf.length + ' wireframes have EMPTY content'); }
+// Check entity attribute types
+const badAttrs = (analysis.entities||[]).flatMap(e => (e.attributes||[]).filter(a => !a.type).map(a => e.name+'.'+a.name));
+if (badAttrs.length > 0) { fails.push(['attr.type', 0, 1]); console.error('FAIL: attributes without type: ' + badAttrs.join(', ')); }
+// AC-15: Validation rules format
+const badRules = (spec.validations||[]).filter(v => v.rules && !Array.isArray(v.rules));
+if (badRules.length > 0) { fails.push(['AC-15: rules not array', badRules.length, 0]); console.error('FAIL: AC-15: ' + badRules.length + ' validations have rules as string'); }
+// AC-16: Messages must have message field
+const noMsg = (spec.messages||[]).filter(m => !m.message);
+if (noMsg.length > 0) { fails.push(['AC-16: message missing', noMsg.length, 0]); console.error('FAIL: AC-16: ' + noMsg.length + ' messages missing message field'); }
+// AC-17: Gherkin content structure
+if (Array.isArray(spec.gherkinScenarios)) { const badG = spec.gherkinScenarios.filter(g => !g.feature || !Array.isArray(g.scenarios)); if (badG.length > 0) { fails.push(['AC-17: gherkin content', badG.length, 0]); console.error('FAIL: AC-17: ' + badG.length + ' gherkin entries invalid'); } }
+if (fails.length > 0) { console.error('BLOCKING: ' + fails.length + ' acceptance criteria failed'); process.exit(1); }
+console.log('PASS: All acceptance criteria met');
+" "$MODULE_JSON"
+```
+
+IF this check FAILS:
+- Identify which criteria failed from the output
+- Fix the corresponding data (re-read from conversation context, re-generate, or re-run the step that produces it)
+- Re-write to feature.json
+- Re-run the POST-CHECK until PASS
+
+> **WHY:** Step-03b now writes wireframes intermediately (since fix), but step-03d section 11 does the FULL write.
+> This bash POST-CHECK is the ultimate safety net — it reads the REAL file and verifies ALL acceptance criteria.
+> The pseudocode checks (section 9a) catch issues early; this bash check catches anything the model missed.
+
+---
+
+### 11-bis. Deploy Incremental Interactive HTML (MANDATORY)
+
+> **After each module is specified, deploy/update the interactive HTML document with all available data.**
+> The client can review completed modules while the next module is being specified.
+
+See [references/validate-incremental-html.md](../references/validate-incremental-html.md) for the complete deployment process (read sources, build FEATURE_DATA, replace placeholders, write HTML).
+
+Uses the **same mapping** as step-05b-deploy.md — only difference is `moduleSpecs` only includes completed modules.
+
+---
+
+### 12. Loop Decision
+
+> **TEAM AGENT MODE: SKIP THIS ENTIRE SECTION.** If you are a team agent (your prompt contains `PROPOSE & REVIEW`), go directly to **section 12-bis** below. Section 12 is for inline mode ONLY. You MUST NOT load the next step or advance the module loop — the team lead handles orchestration.
+
+**INLINE MODE ONLY (main conversation):**
+
+```
+ba-writer.advanceModuleLoop({feature_id})
+→ Increments currentModuleIndex
+→ Updates completedModules
+→ Updates currentModule
+
+IF currentModuleIndex < moduleOrder.length:
+  Display: "→ Module suivant: {nextModule} ({currentModuleIndex + 1}/{total})"
+  Load: steps/step-03a1-setup.md
+
+IF currentModuleIndex >= moduleOrder.length:
+  ba-writer.updateStatus({feature_id}, "specified")
+
+  // CHECKPOINT: Save progress BEFORE transition (protects against context exhaustion)
+  ba-writer.enrichSection({
+    featureId: {feature_id},
+    section: "metadata.workflow",
+    data: {
+      lastCompletedStep: "step-03d-validate",
+      allModulesSpecified: true
+    }
+  })
+
+  // CHECK: Are we in project mode (multi-application)?
+  IF workflow.mode === "project":
+    // APPLICATION LOOP: Check if more applications remain
+    projectFeature = ba-reader.findProjectFeature()
+    ba-writer.updateApplicationStatus(projectId, currentApp.code, "specified")
+    ba-writer.advanceApplicationLoop(projectId)
+
+    IF projectFeature.metadata.workflow.currentApplicationIndex + 1 < projectFeature.metadata.workflow.applicationOrder.length:
+      nextApp = projectFeature.metadata.workflow.applicationOrder[currentApplicationIndex + 1]
+      Display: "═══ Application {currentApp.name} terminée ({N}/{total}). Prochaine : {nextApp} ═══"
+      Display: "⚠ NE PAS lancer /ralph-loop — d'autres applications et la consolidation sont encore nécessaires."
+      // Load step-02 for next application's module decomposition
+      Load: steps/step-02-decomposition.md
+    ELSE:
+      Display: "═══ Toutes les applications spécifiées! Passage à la consolidation cross-application... ═══"
+      Display: "⚠ NE PAS lancer /ralph-loop — la consolidation et le handoff sont encore nécessaires."
+      ba-writer.updateStatus(projectId, "specified")
+      Load: steps/step-04a-collect.md
+  ELSE:
+    // SINGLE APPLICATION: Current behavior
+    Display: "═══ Tous les modules spécifiés! Passage à la consolidation (04a→04c) puis handoff (05a→05c)... ═══"
+    Display: "⚠ NE PAS lancer /ralph-loop — la consolidation et le handoff sont encore nécessaires."
+
+    // MANDATORY TRANSITION — DO NOT STOP HERE
+    // If context is near exhaustion, the checkpoint above ensures the user
+    // can resume with /business-analyse and it will route to step-04a automatically.
+    Load: steps/step-04a-collect.md
+```
+
+---
+
+## 12-bis. Agent Mode: Send Proposal to Team Lead
+
+**IF running as TEAM AGENT (autonomous mode — detected in step-03a1 MODE DETECTION):**
+
+After all validation checks PASS and module feature.json is written:
+
+### A. Build Structured Summary
+
+Read the written module feature.json via ba-reader and build:
+
+```
+## Module: {moduleCode} — Specification Summary
+
+### Entities ({count})
+| Entity | Key Attributes | Relationships |
+|--------|---------------|---------------|
+| {name} | {attr1} ({type}), {attr2} ({type}), ... | → {relatedEntity} (FK) |
+
+### Sections ({count})
+| Section | Description | Resources |
+|---------|-------------|-----------|
+| {code} | {description} | {resource1}, {resource2}, ... |
+
+### Use Cases ({count})
+- {UC-ID}: {title}
+
+### Business Rules ({count})
+- {BR-ID}: {title} ({category})
+
+### Permissions
+- {permission.path}: {roles with access}
+
+### Seed Data Counts
+- Modules: {n} | Sections: {n} | Resources: {n}
+- Translations: {n} | Permissions: {n} | Role mappings: {n} | Constants: {n}
+
+### Wireframes
+- {section}: {1-line description of layout and key components}
+```
+
+### B. Send Proposal
+
+```
+SendMessage({
+  type: "message",
+  recipient: "team-lead",
+  content: "PROPOSAL_READY:{moduleCode}\n\n{structured_summary}",
+  summary: "{moduleCode} specification ready for review"
+})
+```
+
+### C. Wait for Response
+
+**WAIT** for team lead to respond:
+
+- **`APPROVED:{moduleCode}`** → Proceed to step D (Module Complete)
+- **`REVISION:{moduleCode}\n{feedback}`** → Apply changes:
+  1. Read the feedback
+  2. Update the relevant sections in feature.json via ba-writer
+  3. Re-run validation checks (section 9)
+  4. Re-build summary and send new `PROPOSAL_READY`
+  5. Max **3 revision cycles** — after 3, send `ERROR:{moduleCode}:Max revisions reached` and stop
+
+### D. Module Complete
+
+After approval:
+
+```
+ba-writer.updateModuleStatus({feature_id}, {moduleCode}, "specified")
+```
+
+```
+SendMessage({
+  type: "message",
+  recipient: "team-lead",
+  content: "MODULE_COMPLETE:{moduleCode}",
+  summary: "{moduleCode} module complete"
+})
+```
+
+Then WAIT — the team lead will send you a `shutdown_request`.
+Do NOT loop to next module. The team lead handles module ordering and will spawn a new agent for the next module.
+
+### E. Shutdown (MANDATORY)
+
+When you receive a `shutdown_request` from the team lead:
+
+```
+SendMessage({
+  type: "shutdown_response",
+  request_id: "{requestId from the shutdown_request message}",
+  approve: true
+})
+```
+
+This terminates your process. Do NOT output any text after `shutdown_response`.
+
+**Sequence reminder:** APPROVED → MODULE_COMPLETE → shutdown_request → shutdown_response (4 separate steps, never skip).
+
+---
+
+## SELF-VERIFICATION (MANDATORY before loading next step)
+
+Before proceeding to step-03a (next module) or step-04 (consolidation), VERIFY:
+
+1. **Module feature.json has specification section** with ALL sub-sections (actors, useCases, functionalRequirements, permissionMatrix, navigation, seedDataCore, gherkinScenarios, validations, messages, lifeCycles, apiEndpoints, i18nKeys, sections, uiWireframes)
+2. **Module feature.json status** = "specified" (set by section 11)
+3. **Master modules[].status** for this module = "specified" (set by section 11)
+4. **ba-interactive.html exists** and includes data for this module (set by section 11-bis)
+
+**IF any check fails → FIX before proceeding.** Do NOT advance the module loop with incomplete data.
+
+---
+
+## SINGLE-MODULE MODE
+
+When only 1 module in moduleOrder:
+
+- Same full validation flow
+- Skip cross-module references (section 5)
+- After validation → direct to step-04a-collect.md (auto-consolidation)
+
+---
+
+## RESUME SUPPORT
+
+IF the step is resumed (e.g., after interruption):
+
+```
+Read metadata.workflow.currentModuleIndex
+Read metadata.workflow.currentModule
+→ Resume at the correct module
+→ If module has partial data, show what exists and continue from there
+```
+
+---
+
+## CONTEXT MANAGEMENT
+
+| Loaded | Not Loaded |
+|--------|------------|
+| _shared.md | Other module feature.json (full) |
+| feature-schema.json | Questionnaires not needed |
+| Questionnaires for this module | Templates |
+| Completed modules summary (100 lines max) | React viewer files |
+
+---
+
+## STATE VARIABLES
+
+| Variable | Description |
+|----------|-------------|
+| `{currentModule}` | Current module being specified |
+| `{currentModuleIndex}` | Index in moduleOrder |
+| `{completedModules}` | Array of already specified module codes |
+| `{module_feature_id}` | Feature ID for the module-level feature.json |
+
+---
+
+## SUCCESS METRICS
+
+- Module specification validation PASS
+- Per-module validation results recorded in feature.json
+- Client confirmed module specification
+- Module feature.json written with all sections
+- Master updated with module status
+- Interactive HTML deployed/updated with completed module data (incremental)
+- Loop advanced correctly or consolidation phase initiated
+
+---
+
+## FAILURE MODES
+
+- Validation FAIL → offer auto-correct / manual revise / force
+- Client rejects → revise specific sections via step-03a or step-03b
+- Cross-module reference error → clarify with client
+- Interruption → resumable via workflow state
+
+---
+
+## NEXT STEP
+
+Conditional:
+
+- IF modules remaining → Load `steps/step-03a1-setup.md`
+- IF all modules done → Load `steps/step-04a-collect.md`