create-byan-agent 2.3.6 → 2.4.1

package/CHANGELOG.md

@@ -7,6 +7,180 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ---
 
+## [2.4.0] - 2026-02-11
+
+### ✨ Added - Claude Code Native Agent Integration
+
+**New Feature: BYAN agents natively integrated with Claude Code**
+
+Claude Code uses `.claude/CLAUDE.md` and `.claude/rules/*.md` for project memory.
+Yanstaller now creates this structure automatically when Claude is detected.
+
+**Files Created in User Project:**
+```
+.claude/
+  CLAUDE.md                      # Main project memory with Hermes entry point
+  rules/
+    hermes-dispatcher.md         # Hermes commands, routing rules, pipelines
+    byan-agents.md               # 35+ agents across 5 modules (tables)
+    merise-agile.md              # Methodology, mantras, dev cycle, test levels
+```
+
+**Hermes Always Included:**
+- Hermes dispatcher is the universal entry point on ALL Claude Code projects
+- `CLAUDE.md` references Hermes and links to rules via `@.claude/rules/` imports
+- Claude Code auto-loads all `.claude/rules/*.md` at session start
+
+**How It Works:**
+- User runs `npx create-byan-agent` and selects Claude Code platform
+- Yanstaller detects Claude and installs `.claude/` structure
+- Claude Code reads `CLAUDE.md` + all `rules/*.md` at every session start
+- User asks "quel agent pour mon projet?" → Claude knows Hermes and all agents
+- No MCP server needed for agent knowledge (native Claude Code memory)
+
+**Verification:**
+- Installation checks: CLAUDE.md, rules/ directory, hermes-dispatcher.md
+- Updated success message with Claude-specific activation instructions
+
+**Based on Claude Code SDK:**
+- Uses official `.claude/rules/*.md` modular rules system
+- Uses `@path` import syntax for cross-referencing
+- Rules auto-loaded per session (no manual configuration)
+
+---
+
+## [2.3.8] - 2026-02-11
+
+### 🐛 Fixed - Windows 11 + Claude Code Compatibility
+
+**Cross-Platform CLI Execution:**
+- Replaced `execSync` with `spawnSync` in Phase 2 chat (no shell = no character interpretation)
+- On Unix: args passed directly (no `@`, `$`, `%` interpretation)
+- On Windows: `shell: true` for `.cmd` file support, stdin for long prompts
+- Removed all bash-only syntax: `$(cat ...)`, `2>/dev/null`, single-quote escaping
+
+**Version Display Fix:**
+- `BYAN_VERSION` now reads from `package.json` (was hardcoded as `2.3.0`)
+- Banner now shows correct installed version
+
+**Claude-Aware Model Selection:**
+- When Claude platform selected, model switches from `gpt-5-mini` to `claude-haiku-4.5`
+- `generateDefaultConfig()` now accepts `selectedPlatform` parameter
+- Display: "Model adapte: claude-haiku-4.5 (plateforme: Claude)"
+
+**Claude Auth Improvements:**
+- Login command fixed: `claude login` (was `claude auth`)
+- Shows 3 connection methods on failure:
+  1. `claude login` (OAuth)
+  2. `export ANTHROPIC_API_KEY=sk-ant-...`
+  3. `/login` dans Claude Code
+- Auth error detection in Phase 2 chat with specific guidance
+
+**Files Modified:**
+- `install/lib/phase2-chat.js`: New `runCliCommand()` helper, cross-platform `sendChatMessage()`
+- `install/bin/create-byan-agent-v2.js`: `spawnSync` imports, dynamic version, platform-aware model
+
+---
+
+## [2.3.7] - 2026-02-11
+
+### 🐛 Fixed - Codex Prompt Escaping
+
+**Issue:** Bash interpreted `@hermes` as shell command in Codex prompts
+- Phase 2 chat with Codex failed with `/bin/bash: ligne 1: @hermes : commande introuvable`
+- Root cause: `echo "${prompt}"` with double quotes allowed bash to interpret special characters
+- The Hermes documentation contains `@hermes` examples, which bash tried to execute
+
+**Fix:**
+- Changed `install/lib/phase2-chat.js`: Use single quotes for Codex echo command
+- Escape single quotes within content: `replace(/'/g, "'\\''")` 
+- Prevents bash from interpreting `@`, `$`, backticks, and other special chars
+- Copilot and Claude already working (direct arguments, not shell expansion)
+
+**Impact:** Yanstaller Phase 2 now works correctly with Codex platform
+
+---
+
+## [2.3.6] - 2026-02-11
+
+### ✨ Enhanced - Hermes in Yanstaller Knowledge Base
+
+**Feature:** Yanstaller Phase 2 now knows complete BYAN ecosystem
+- Added full Hermes documentation to Phase 2 system prompt (~1500 tokens)
+- Documents all 35+ agents across 5 modules (Core, BMM, BMB, CIS, TEA)
+- Includes 7 predefined workflows with agent chains
+- Explains smart routing capabilities and pipelines
+- Yanstaller recommends Hermes as universal entry point
+
+**Knowledge Expansion:**
+- Before: ~300 tokens (basic user profile)
+- After: ~1500 tokens (complete ecosystem)
+- 5x increase in contextual intelligence
+
+**Impact:** Users asking about Hermes or agents get intelligent, informed responses
+
+---
+
+## [2.3.5] - 2026-02-11
+
+### 🐛 Fixed - Multi-Platform CLI Commands
+
+**Issue:** Codex and Claude commands were incorrect
+- Codex: `codex -p "prompt"` doesn't exist (wrong flag)
+- Claude: `claude -p "prompt" --no-input` used wrong flags
+- Both failed during Phase 2 chat in Yanstaller
+
+**Fix:**
+- `install/lib/phase2-chat.js`:
+  - Codex: Changed to `codex exec` with stdin (line 175)
+  - Claude: Changed to `claude -p` print mode (line 184)
+  - Copilot: Already correct with `-p` flag
+
+**Verified:**
+- Copilot: `copilot -p "prompt" -s` ✓
+- Codex: `echo "prompt" | codex exec` ✓
+- Claude: `claude -p "prompt"` ✓
+
+---
+
+## [2.3.4] - 2026-02-11
+
+### 🐛 Fixed - Yanstaller Phase 2 Config Bug
+
+**Issue:** ReferenceError during Phase 2 chat initialization
+- Error: `Cannot access 'config' before initialization`
+- Root cause: `config` variable used at line 737 but defined at line 817
+- Phase 2 chat failed to start
+
+**Fix:**
+- `install/bin/create-byan-agent-v2.js`: Moved config prompt to Phase 1.5 (before Phase 2 chat)
+- Config now collected at lines 714-729 (userName, language)
+- Passed to `launchPhase2Chat()` as parameters
+- Removed duplicate config definition
+
+**Impact:** Yanstaller Phase 2 now starts correctly with all platforms
+
+---
+
+## [2.3.3] - 2026-02-10
+
+### 🐛 Fixed - GitHub Repository URLs
+
+**Issue:** README and package.json pointed to wrong GitHub repository
+- Old URL: `github.com/yannsix/byan-v2` (incorrect)
+- Correct URL: `github.com/Yan-Acadenice/BYAN`
+
+**Changes:**
+- `README.md`: Updated 10 GitHub link occurrences
+- `package.json`: Updated repository URLs (3 occurrences)
+  - repository.url
+  - bugs.url
+  - homepage
+
+**Impact:** Users can now find correct repository and report issues properly
+
+---
+
 ## [2.3.2] - 2026-02-10
 
 ### 🏛️ Added - Hermes Universal Dispatcher

package/install/bin/create-byan-agent-v2.js

@@ -3,7 +3,7 @@
 const fs = require('fs-extra');
 const path = require('path');
 const os = require('os');
-const { execSync } = require('child_process');
+const { execSync, spawnSync } = require('child_process');
 const { program } = require('commander');
 const inquirer = require('inquirer');
 const chalk = require('chalk');
@@ -15,7 +15,7 @@ const { getDomainQuestions, buildPhase2Prompt } = require('../lib/domain-questio
 const { generateProjectAgentsDoc } = require('../lib/project-agents-generator');
 const { launchPhase2Chat, generateDefaultConfig } = require('../lib/phase2-chat');
 
-const BYAN_VERSION = '2.3.0';
+const BYAN_VERSION = require('../../package.json').version;
 
 // ASCII Art Banner
 const banner = `
@@ -470,29 +470,41 @@ async function install() {
     const promptFile = path.join(projectRoot, '.yanstaller-prompt.tmp');
     await fs.writeFile(promptFile, interviewPrompt, 'utf8');
     
-    // Detect which CLI to use (non-interactive mode)
-    // Copilot: --agent loads from .github/agents/, -p for non-interactive, -s for silent
-    // Codex: exec subcommand for non-interactive, -m for model
-    // Claude: -p for non-interactive prompt
-    let agentCommand = null;
-    if (detectedPlatforms.copilot) {
-      agentCommand = `copilot --agent=bmad-agent-yanstaller-interview -p "$(cat '${promptFile}')" --model ${interviewComplexity} -s`;
-    } else if (detectedPlatforms.codex) {
-      agentCommand = `codex exec "$(cat '${promptFile}')"`;
-    } else if (detectedPlatforms.claude) {
-      agentCommand = `claude -p "$(cat '${promptFile}')"`;
-    }
+    // Cross-platform CLI execution using spawnSync (no shell = no escaping issues)
+    const isWindows = process.platform === 'win32';
+    const promptContent = await fs.readFile(promptFile, 'utf8');
+    let hasAgent = false;
     
-    if (agentCommand) {
+    if (detectedPlatforms.copilot || detectedPlatforms.codex || detectedPlatforms.claude) {
+      hasAgent = true;
       const agentSpinner = ora(`Analysing with yanstaller agent (${interviewComplexity})...`).start();
       
       try {
-        const result = execSync(agentCommand, {
+        const spawnOpts = {
           encoding: 'utf8',
           cwd: projectRoot,
           timeout: 120000,
+          maxBuffer: 1024 * 1024,
           stdio: ['pipe', 'pipe', 'pipe']
-        });
+        };
+        if (isWindows) spawnOpts.shell = true;
+        
+        let res;
+        if (detectedPlatforms.copilot) {
+          res = spawnSync('copilot', [
+            '--agent=bmad-agent-yanstaller-interview',
+            '-p', promptContent,
+            '--model', interviewComplexity,
+            '-s'
+          ], spawnOpts);
+        } else if (detectedPlatforms.codex) {
+          res = spawnSync('codex', ['exec', promptContent], spawnOpts);
+        } else if (detectedPlatforms.claude) {
+          res = spawnSync('claude', ['-p', promptContent], spawnOpts);
+        }
+        
+        if (res.error) throw res.error;
+        const result = (res.stdout || '').toString();
         
         agentSpinner.succeed(`Analysis complete (model: ${interviewComplexity})`);
         
@@ -568,7 +580,7 @@ async function install() {
         console.log(chalk.green(`  Essential agents: ${interviewResults.agents.essential.join(', ')}`));
         console.log(chalk.green(`  Model: ${interviewResults.recommended_model}`));
       }
-    } else {
+    } else if (!hasAgent) {
       // No AI platform detected - build from interview data
       interviewResults = {
         platforms: ['copilot'],
@@ -663,6 +675,15 @@ async function install() {
       installMode = 'auto';
     }
     
+    // Update recommended model based on selected platform
+    if (selectedPlatform === 'claude' && interviewResults) {
+      const oldModel = interviewResults.recommended_model || '';
+      if (oldModel.includes('gpt')) {
+        interviewResults.recommended_model = 'claude-haiku-4.5';
+        console.log(chalk.cyan(`  🧠 Model adapté: claude-haiku-4.5 (plateforme: Claude)`));
+      }
+    }
+    
     // NEW: Verify authentication for selected platform
     if (selectedPlatform && installMode === 'custom') {
       console.log('');
@@ -680,7 +701,7 @@ async function install() {
           loginCmd = 'codex login';
         } else if (selectedPlatform === 'claude') {
           checkCmd = 'claude --version';
-          loginCmd = 'claude auth';
+          loginCmd = 'claude login';
         }
         
         const result = execSync(checkCmd, { encoding: 'utf8', timeout: 5000, stdio: 'pipe' });
@@ -689,6 +710,10 @@ async function install() {
         console.log('');
         console.log(chalk.yellow(`⚠️  ${selectedPlatform} n'est pas authentifié ou non disponible`));
         console.log(chalk.gray(`   Pour vous connecter, exécutez: ${chalk.cyan(loginCmd)}`));
+        if (selectedPlatform === 'claude') {
+          console.log(chalk.gray(`   Alternative: ${chalk.cyan('export ANTHROPIC_API_KEY=sk-ant-...')}`));
+          console.log(chalk.gray(`   Ou dans Claude Code: ${chalk.cyan('/login')}`));
+        }
         console.log('');
         
         const { continueAnyway } = await inquirer.prompt([{
@@ -766,13 +791,13 @@ async function install() {
         }]);
         
         if (useFallback) {
-          phase2Results = generateDefaultConfig(interviewAnswers, detectedPlatforms);
+          phase2Results = generateDefaultConfig(interviewAnswers, detectedPlatforms, selectedPlatform);
         }
       }
     } else if (enterPhase2 === 'auto') {
       // Use default configuration
       const autoSpinner = ora('Generating default configuration...').start();
-      phase2Results = generateDefaultConfig(interviewAnswers, detectedPlatforms);
+      phase2Results = generateDefaultConfig(interviewAnswers, detectedPlatforms, selectedPlatform);
       autoSpinner.succeed('Configuration generated');
     }
     
@@ -918,6 +943,9 @@ async function install() {
   await fs.ensureDir(path.join(byanDir, '_memory'));
   await fs.ensureDir(path.join(byanDir, '_output'));
   await fs.ensureDir(githubAgentsDir);
+  if (detectedPlatforms.claude || platform === 'claude' || platform === 'all') {
+    await fs.ensureDir(path.join(projectRoot, '.claude', 'rules'));
+  }
   
   installSpinner.succeed('Directory structure created');
   
@@ -978,6 +1006,19 @@ async function install() {
       copySpinner.warn(`⚠ GitHub agents source not found: ${githubAgentsSource}`);
     }
     
+    // Copy .claude/ files for Claude Code integration (always includes Hermes)
+    if (detectedPlatforms.claude || platform === 'claude' || platform === 'all') {
+      const claudeSource = path.join(templateDir, '.claude');
+      const claudeDest = path.join(projectRoot, '.claude');
+      
+      if (await fs.pathExists(claudeSource)) {
+        await fs.ensureDir(path.join(claudeDest, 'rules'));
+        await fs.copy(claudeSource, claudeDest, { overwrite: true });
+        copySpinner.text = 'Copied Claude Code rules + Hermes dispatcher...';
+        console.log(chalk.green(`  ✓ Claude Code: CLAUDE.md + rules/ (Hermes, agents, methodology)`));
+      }
+    }
+    
     copySpinner.succeed('BYAN platform files installed');
   } catch (error) {
     copySpinner.fail('Error copying files');
@@ -1033,6 +1074,15 @@ async function install() {
     { name: 'GitHub agents dir', path: githubAgentsDir }
   ];
   
+  // Add Claude Code checks if installed
+  if (detectedPlatforms.claude || platform === 'claude' || platform === 'all') {
+    checks.push(
+      { name: 'Claude CLAUDE.md', path: path.join(projectRoot, '.claude', 'CLAUDE.md') },
+      { name: 'Claude rules/', path: path.join(projectRoot, '.claude', 'rules') },
+      { name: 'Hermes dispatcher rule', path: path.join(projectRoot, '.claude', 'rules', 'hermes-dispatcher.md') }
+    );
+  }
+  
   // Add v2.0 checks if installed
   if (v2Installed) {
     checks.push(
@@ -1122,7 +1172,10 @@ async function install() {
     console.log('   Open VSCode Command Palette (Ctrl+Shift+P)');
     console.log('   Type: \'Activate BYAN Agent\'');
   } else if (platform === 'claude') {
-    console.log(`   ${chalk.blue('claude chat --agent byan')}`);
+    console.log(`   ${chalk.blue('claude')}`);
+    console.log(`   Hermes est integre via ${chalk.cyan('.claude/CLAUDE.md')}`);
+    console.log(`   Demande: ${chalk.cyan('"quel agent pour mon projet?"')} → Hermes repond`);
+    console.log(`   Regles: ${chalk.cyan('.claude/rules/')} (hermes, agents, methodologie)`);
   } else {
     console.log('   Follow your platform\'s agent activation procedure');
   }

package/install/lib/phase2-chat.js

@@ -5,13 +5,43 @@
  * using copilot/codex CLI for AI responses.
  */
 
-const { execSync } = require('child_process');
+const { spawnSync } = require('child_process');
 const path = require('path');
 const fs = require('fs-extra');
 const inquirer = require('inquirer');
 const chalk = require('chalk');
 const ora = require('ora');
 
+/**
+ * Cross-platform CLI command execution
+ * Uses spawnSync without shell on Unix (no character interpretation)
+ * Uses shell on Windows for .cmd file support
+ */
+function runCliCommand(cmd, args, cwd, stdinInput) {
+  const isWindows = process.platform === 'win32';
+  const opts = {
+    encoding: 'utf8',
+    cwd: cwd,
+    timeout: 60000,
+    maxBuffer: 1024 * 1024,
+    stdio: ['pipe', 'pipe', 'pipe']
+  };
+  
+  if (stdinInput) opts.input = stdinInput;
+  if (isWindows) opts.shell = true;
+  
+  const res = spawnSync(cmd, args, opts);
+  if (res.error) throw res.error;
+  
+  // Check for non-zero exit code with stderr
+  if (res.status !== 0 && res.stderr) {
+    const stderr = res.stderr.toString().trim();
+    if (stderr) throw new Error(stderr);
+  }
+  
+  return (res.stdout || '').toString();
+}
+
 /**
  * Build the initial context prompt from Phase 1 answers
  */
@@ -144,7 +174,6 @@ Quand l'utilisateur dit "finaliser", "terminer" ou "c'est bon", génère la conf
  * @returns {Promise<string>} AI response
  */
 async function sendChatMessage(message, systemContext, conversationHistory, selectedPlatform, projectRoot) {
-  // Build the full prompt with context and history
   const fullPrompt = `${systemContext}
 
 ## Historique de conversation:
@@ -160,44 +189,37 @@ Continue la conversation pour comprendre le projet et personnaliser les agents.`
   let result = '';
   
   try {
-    const escaped = fullPrompt.replace(/"/g, '\\"').replace(/\n/g, '\\n');
-    
     if (selectedPlatform === 'copilot') {
-      // Use copilot with single prompt mode (-s for silent, no interactive)
-      result = execSync(`copilot -p "${escaped}" -s 2>/dev/null`, {
-        encoding: 'utf8',
-        cwd: projectRoot,
-        timeout: 60000,
-        maxBuffer: 1024 * 1024
-      });
+      result = runCliCommand('copilot', ['-p', fullPrompt, '-s'], projectRoot);
     } else if (selectedPlatform === 'codex') {
-      // Use codex exec with stdin for long prompts (avoids escaping issues)
-      result = execSync(`echo "${escaped}" | codex exec 2>/dev/null`, {
-        encoding: 'utf8',
-        cwd: projectRoot,
-        timeout: 60000,
-        maxBuffer: 1024 * 1024,
-        shell: '/bin/bash'
-      });
+      // Codex takes prompt as argument to exec command
+      result = runCliCommand('codex', ['exec', fullPrompt], projectRoot);
     } else if (selectedPlatform === 'claude') {
-      // Claude takes prompt as argument with -p flag for print mode
-      result = execSync(`claude -p "${escaped}" 2>/dev/null`, {
-        encoding: 'utf8',
-        cwd: projectRoot,
-        timeout: 60000,
-        maxBuffer: 1024 * 1024
-      });
+      result = runCliCommand('claude', ['-p', fullPrompt], projectRoot);
     } else {
       throw new Error(`Platform not supported: ${selectedPlatform}`);
     }
   } catch (error) {
+    const errMsg = (error.message || '').toLowerCase();
     console.error(chalk.red(`\n❌ Erreur ${selectedPlatform}: ${error.message}`));
+    
+    // Platform-specific login guidance
+    if (selectedPlatform === 'claude' && (errMsg.includes('auth') || errMsg.includes('api') || errMsg.includes('key') || errMsg.includes('login') || errMsg.includes('401'))) {
+      console.log('');
+      console.log(chalk.yellow('  💡 Pour se connecter à Claude Code:'));
+      console.log(chalk.cyan('     1. claude login'));
+      console.log(chalk.gray('     2. ou: export ANTHROPIC_API_KEY=sk-ant-...'));
+      console.log(chalk.gray('     3. ou dans Claude Code: /login'));
+    } else if (selectedPlatform === 'copilot') {
+      console.log(chalk.gray('     → gh auth login'));
+    } else if (selectedPlatform === 'codex') {
+      console.log(chalk.gray('     → codex login'));
+    }
+    
     result = `Désolé, erreur de communication avec ${selectedPlatform}. Réessayez ou tapez "skip".`;
   }
   
-  // Clean up the response (remove ANSI codes, extra whitespace)
   result = result.replace(/\x1b\[[0-9;]*m/g, '').trim();
-  
   return result;
 }
 
@@ -373,7 +395,7 @@ IMPORTANT: L'utilisateur veut finaliser. Génère maintenant la configuration JS
 /**
  * Fallback: Generate default config based on Phase 1 answers only
  */
-function generateDefaultConfig(interviewAnswers, detectedPlatforms) {
+function generateDefaultConfig(interviewAnswers, detectedPlatforms, selectedPlatform) {
   const domain = interviewAnswers.domain;
   const quality = interviewAnswers.quality;
   
@@ -477,9 +499,13 @@ function generateDefaultConfig(interviewAnswers, detectedPlatforms) {
     { from: 'architect', to: 'dev', type: 'triggers', description: 'Architecture triggers implementation' }
   ];
   
-  // Set model based on quality
-  config.recommendedModel = quality === 'critical' ? 'claude-sonnet-4' : 
-                            quality === 'production' ? 'gpt-5.1-codex' : 'gpt-5-mini';
+  // Set model based on quality and selected platform
+  if (selectedPlatform === 'claude') {
+    config.recommendedModel = quality === 'critical' ? 'claude-sonnet-4' : 'claude-haiku-4.5';
+  } else {
+    config.recommendedModel = quality === 'critical' ? 'claude-sonnet-4' : 
+                              quality === 'production' ? 'gpt-5.1-codex' : 'gpt-5-mini';
+  }
   
   config.rationale = `Configuration par défaut pour projet ${domain}. Personnalisez via la conversation Phase 2 pour plus de précision.`;
   

package/install/templates/.claude/CLAUDE.md

@@ -0,0 +1,44 @@
+# BYAN - Builder of YAN
+
+> Projet propulse par BYAN (Merise Agile + TDD + 64 Mantras)
+> Installer: `npx create-byan-agent`
+> GitHub: https://github.com/Yan-Acadenice/BYAN
+
+## Hermes - Dispatcher Universel
+
+**Hermes est le point d'entree universel de ton ecosysteme BYAN.**
+Avant de chercher un agent specifique, demande a Hermes. Il connait tous les agents,
+workflows et contextes, et te route vers le bon specialiste.
+
+Pour invoquer Hermes, tape: `@hermes` ou demande simplement "quel agent pour [ta tache]?"
+
+Voir @.claude/rules/hermes-dispatcher.md pour les commandes Hermes.
+
+## Architecture BYAN
+
+```
+{project-root}/
+  _byan/              # Plateforme BYAN
+    _config/           # Manifestes (agents, workflows, tasks)
+    bmb/               # Module Builder (BYAN, agents, workflows)
+    _memory/           # Memoire persistante des agents
+    _output/           # Artefacts generes
+  .claude/             # Integration Claude Code
+    CLAUDE.md          # Ce fichier (instructions projet)
+    rules/             # Regles modulaires par domaine
+  .github/agents/      # Agents Copilot CLI (si installe)
+```
+
+## Regles de Code
+
+- Pas d'emojis dans le code, commits, ou specs techniques (Mantra IA-23)
+- Code auto-documente, commentaires uniquement pour le POURQUOI (Mantra IA-24)
+- Format commits: `type: description` (feat, fix, docs, refactor, test, chore)
+- Simplicite d'abord - Rasoir d'Ockham (Mantra #37)
+- Challenge Before Confirm - Valider avant d'accepter (Mantra IA-16)
+
+## Commandes Utiles
+
+- `@hermes` → Dispatcher universel (recommandations, routage, pipelines)
+- Agent disponibles: voir @.claude/rules/byan-agents.md
+- Methodologie: voir @.claude/rules/merise-agile.md

package/install/templates/.claude/rules/byan-agents.md

@@ -0,0 +1,72 @@
+# Agents BYAN - Ecosysteme Complet
+
+## Core Module (Foundation)
+
+| Agent | Persona | Role |
+|-------|---------|------|
+| **hermes** | Dispatcher | Routeur universel, point d'entree |
+| **bmad-master** | Orchestrateur | Execute workflows et tasks |
+| **yanstaller** | Installeur | Installation intelligente BYAN |
+| **expert-merise-agile** | Expert | Conception Merise Agile + MCD/MCT |
+
+## BMB Module (Builders)
+
+| Agent | Persona | Role |
+|-------|---------|------|
+| **byan** | Builder | Createur d'agents via interview (12 questions, 64 mantras) |
+| **agent-builder** | Constructeur | Expert en construction d'agents |
+| **marc** | Specialiste | Integration GitHub Copilot |
+| **rachid** | Specialiste | Deploiement NPM/NPX |
+| **carmack** | Optimiseur | Optimisation tokens |
+| **patnote** | Gestionnaire | Mises a jour et conflits |
+
+## BMM Module (SDLC - Software Development Lifecycle)
+
+| Agent | Persona | Role |
+|-------|---------|------|
+| **analyst** | Mary | Analyse business, etude de marche, brief |
+| **architect** | Winston | Design systeme, tech stack, architecture |
+| **dev** | Amelia | Implementation, coding, ultra-succincte |
+| **pm** | John | Product management, PRD, roadmap |
+| **sm** | Bob | Scrum master, sprint planning, backlog |
+| **quinn** | Quinn | QA engineer, tests, couverture |
+| **tech-writer** | Paige | Documentation, guides, clarity |
+| **ux-designer** | Sally | UX/UI design, empathie utilisateur |
+| **quick-flow-solo-dev** | Barry | Dev rapide brownfield |
+
+## CIS Module (Creative Innovation & Strategy)
+
+| Agent | Persona | Role |
+|-------|---------|------|
+| **brainstorming-coach** | Carson | Ideation, "YES AND" energy |
+| **creative-problem-solver** | Dr. Quinn | Resolution de problemes |
+| **design-thinking-coach** | Maya | Design thinking |
+| **innovation-strategist** | Victor | Strategie innovation |
+| **presentation-master** | Caravaggio | Presentations, slides |
+| **storyteller** | Sophia | Storytelling, narratives |
+
+## TEA Module (Test Engineering & Architecture)
+
+| Agent | Persona | Role |
+|-------|---------|------|
+| **tea** | Murat | Master test architect (ATDD, NFR, CI/CD) |
+
+## Workflows Cles
+
+| Workflow | Description |
+|----------|-------------|
+| `create-prd` | Creer un Product Requirements Document |
+| `create-architecture` | Concevoir l'architecture technique |
+| `create-epics-and-stories` | Decouper en epics et user stories |
+| `sprint-planning` | Planifier un sprint |
+| `dev-story` | Developper une story |
+| `code-review` | Revoir du code |
+| `quick-spec` | Spec rapide conversationnelle |
+| `quick-dev` | Dev rapide (brownfield) |
+
+## Comment Invoquer un Agent
+
+Dans Claude Code, demande simplement:
+- "Je veux creer une architecture" → Hermes recommande `architect`
+- "Analyse ce projet" → Hermes recommande `analyst`
+- "Cree un nouvel agent" → Hermes recommande `byan`

package/install/templates/.claude/rules/hermes-dispatcher.md

@@ -0,0 +1,53 @@
+# Hermes - Dispatcher Universel BYAN
+
+Hermes est le routeur intelligent de l'ecosysteme BYAN. Il ne fait pas le travail
+lui-meme, il invoque le bon specialiste.
+
+## Commandes Hermes
+
+| Commande | Action |
+|----------|--------|
+| `[LA]` | Lister tous les agents par module |
+| `[LW]` | Lister les workflows disponibles |
+| `[LC]` | Lister les contextes projet |
+| `[REC]` | Recommandation: decris ta tache, Hermes trouve le bon agent |
+| `[PIPE]` | Pipelines multi-agents pour taches complexes |
+| `[?agent]` | Quick help sur un agent sans le charger |
+| `[@agent]` | Invoquer directement un agent |
+| `[HELP]` | Reafficher le menu |
+| `[EXIT]` | Quitter Hermes |
+
+## Routage Intelligent
+
+Quand un utilisateur decrit une tache, Hermes recommande le bon agent:
+
+| Mots-cles | Agent recommande |
+|-----------|------------------|
+| analyser, requirements, brief, etude | analyst (Mary) |
+| architecture, design, tech stack | architect (Winston) |
+| coder, implementer, dev, feature | dev (Amelia) |
+| tester, QA, coverage, bugs | quinn (QA) / tea (Murat) |
+| planifier, sprint, backlog, scrum | sm (Bob) |
+| documenter, guide, readme | tech-writer (Paige) |
+| UX, design, mockup, interface | ux-designer (Sally) |
+| PRD, produit, roadmap, specs | pm (John) |
+| creer agent, workflow, module | byan (Builder) |
+| brainstorm, idees, innovation | brainstorming-coach (Carson) |
+| optimiser, tokens, performance | carmack (Optimizer) |
+
+## Pipelines Predefinies
+
+1. **Feature Complete**: PM → Architect → UX → SM → Dev → Tea
+2. **Idea to Code**: PM → Architect → SM → Quick Flow
+3. **New Agent**: BYAN (handles entire flow)
+4. **Refactoring**: Architect → Dev → Tea
+5. **Bug Fix**: Dev → Quinn
+6. **Documentation**: Analyst → Tech Writer
+7. **Quality Complete**: Tea → Quinn → code-review
+
+## Manifestes
+
+Hermes lit les manifestes CSV a l'execution:
+- `_byan/_config/agent-manifest.csv` - Tous les agents installes
+- `_byan/_config/workflow-manifest.csv` - Tous les workflows
+- `_byan/_config/task-manifest.csv` - Toutes les tasks standalone

package/install/templates/.claude/rules/merise-agile.md

@@ -0,0 +1,48 @@
+# Methodologie Merise Agile + TDD
+
+BYAN utilise la methodologie Merise Agile enrichie de 64 mantras.
+
+## Principes Fondamentaux
+
+1. **Data Dictionary First** (Mantra #33): Definir les entites de donnees avant toute modelisation
+2. **MCD-MCT Cross-Validation** (Mantra #34): Coherence entre modeles de donnees et traitements
+3. **Bottom-Up from User Stories**: Les entites emergent des user stories
+4. **Incremental Design**: Sprint 0 = MCD squelettique, enrichi sprint par sprint
+5. **Test-Driven at All Levels**: Tests conceptuels avant implementation
+
+## Mantras Cles
+
+| ID | Mantra | Application |
+|----|--------|-------------|
+| #37 | Rasoir d'Ockham | Simplicite d'abord, approche MVP |
+| #39 | Consequences | Evaluer avant d'executer |
+| IA-1 | Trust But Verify | Challenger toutes les exigences |
+| IA-16 | Challenge Before Confirm | Jouer l'avocat du diable |
+| IA-23 | No Emoji Pollution | Zero emoji dans code, commits, specs |
+| IA-24 | Clean Code | Auto-documente, commentaires minimaux |
+
+## Cycle de Developpement BYAN
+
+```
+Phase 0: Document Project (brownfield)
+Phase 1: Analyse (Brief → PRD)
+Phase 2: Planning (Architecture → Epics/Stories)
+Phase 3: Solutioning (Sprint Planning)
+Phase 4: Implementation (Dev → Test → Review)
+```
+
+## Niveaux de Test
+
+Priorite (preferer les niveaux bas):
+1. **Unit** > **Integration** > **E2E**
+2. Les tests API sont first-class citizens
+3. Tout nouveau code necessite des tests unitaires
+4. Chemins critiques: tests d'integration
+5. Parcours utilisateur: tests E2E
+
+## Convention Commits
+
+Format: `type: description`
+- Types: `feat`, `fix`, `docs`, `refactor`, `test`, `chore`
+- PAS d'emojis dans les commits
+- Description claire et concise en anglais

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "create-byan-agent",
-  "version": "2.3.6",
+  "version": "2.4.1",
   "description": "BYAN v2.3.2 - Intelligent AI agent ecosystem with Hermes universal dispatcher + Multi-platform support (Copilot CLI, Claude, Codex) + Automatic LLM cost optimization (87.5% savings) + Node 12+ compatible",
   "main": "src/index.js",
   "bin": {