@luquimbo/bi-superpowers 3.2.0 → 4.1.1

package/README.md

@@ -5,62 +5,89 @@
 [![Node.js Version](https://img.shields.io/badge/node-%3E%3D18.0.0-brightgreen)](https://nodejs.org)
 [![Power BI](https://img.shields.io/badge/Power%20BI-Desktop-F2C811?logo=powerbi)](https://powerbi.microsoft.com)
 
-Toolkit open-source para trabajar con **Power BI Desktop** desde tu agente AI favorito. Funciona con los 5 agentes soportados: Claude Code, GitHub Copilot, Codex, Gemini CLI y Kilo Code.
+Toolkit open-source para trabajar con **Power BI Desktop** desde Claude Code, GitHub Copilot, Codex, Gemini CLI y Kilo Code.
 
-> **Estado:** versión inicial. 2 skills y 2 MCP servers para empezar a iterar.
+> **Estado:** `v4.1.0` — **primera versión estable**. Incluye 4 skills (`/bi-start` + 3 especialistas) y 2 MCP servers oficiales de Microsoft. `/report-design` genera PBIR via scripts Node propios (no depende de `pbi-cli-tool` para authoring — Windows + Power BI Desktop sí sigue siendo requisito runtime).
+>
+> _Nota de versionado_: las versiones 2.x, 3.x y 4.0.0 fueron iteraciones mientras el alcance y la arquitectura del plugin se estaban asentando. **Tratá a v4.1.x como si fuera la 1.0** del producto — es el primer release que el autor considera listo para uso real. El numeral se queda en la línea 4.x porque npm es estrictamente monotónico (no se puede bajar el número), pero la madurez del producto arranca acá.
 
 ---
 
-## Quick Start
+## Instalación
+
+Elegí el camino según tu agente principal:
+
+### 👉 Claude Code (recomendado para usuarios de Claude Code)
+
+Dentro de Claude Code, ejecutá:
+
+```
+/plugin marketplace add luquimbo/bi-superpowers
+/plugin install bi-superpowers
+```
+
+Eso instala el plugin completo (4 skills + 2 MCPs) en el perfil de Claude Code. Después podés actualizar con `/plugin update bi-superpowers` directamente desde Claude Code — sin pasar por npm.
+
+Si también querés el plugin local de proyecto (con slash commands `/project-kickoff`, `/pbi-connect`, `/report-design` en un repo específico), corré `super kickoff` dentro del repo — requiere haber instalado la CLI via el otro path (abajo).
+
+### 👉 Otros agentes — Copilot, Codex, Gemini CLI, Kilo Code (o multi-agente)
 
 ```bash
 # 1. Instalá la CLI
 npm install -g @luquimbo/bi-superpowers
 
-# 2. Instalá skills y MCPs en todos tus agentes
+# 2. Instalá skills + MCPs en todos tus agentes
 super install
 ```
 
-Eso es todo. El instalador:
+El instalador:
 
 1. Detecta qué agentes AI tenés instalados (Claude Code, Copilot, Codex, Gemini, Kilo).
 2. Te deja elegir para cuáles instalar (o hace todos con `--all --yes`).
-3. Copia los 2 skills en los directorios correctos de cada agente.
+3. Copia los 4 skills en los directorios correctos de cada agente.
 4. Escribe los 2 MCPs en el config file correcto de cada agente (con el formato/schema que cada uno espera).
 
-Reiniciá tu agente AI, y ya tenés bi-superpowers funcionando.
+Reiniciá tu agente AI y ya tenés bi-superpowers funcionando.
+
+> **¿Por qué dos caminos?** El marketplace de Claude Code es la UX más limpia para usuarios de Claude Code — cero dependencias externas, updates nativos. La CLI via npm cubre los otros 4 agentes (que no tienen un marketplace equivalente) y funciona como denominador común cuando querés las mismas skills + MCPs en múltiples agentes. Los dos caminos son compatibles: podés combinar instalación via marketplace en Claude Code con instalación via CLI para los otros.
 
 ---
 
 ## Qué viene incluido
 
-### 2 Skills
+### 4 Skills
 
 | Skill | Para qué |
 |---|---|
-| `project-kickoff` | Analizá el estado de tu proyecto BI y decidí próximos pasos |
+| `bi-start` | **Arrancar una sesión**: muestra el menú de skills, corre update-check, y (si detecta Power BI Desktop abierto) ofrece conectar vía MCP. Tu entrada por defecto en cada conversación nueva. |
+| `project-kickoff` | Analizá el estado de tu proyecto BI **nuevo** y armá `AGENTS.md` + modelo base |
 | `pbi-connect` | Conectá tu agente AI a Power BI Desktop vía MCP |
+| `report-design` | Generá páginas PBIR vía scripts Node propios (`create-visual.js`, `apply-theme.js`, `validate-pbir.js`), con allowlist de 28 visualTypes nativos |
+
+Si no sabés por dónde empezar, decile a tu agente `/bi-start` — te muestra los 4 skills, chequea updates, y rutea al que necesites. `/project-kickoff` está reservado para cuando arrancás un proyecto BI desde cero.
+
+`report-design` requiere **Windows + Power BI Desktop** (standalone, no la versión de Microsoft Store). Los scripts Node son portables, pero el ciclo de verificación (relaunch, render) necesita Desktop. `pbi-cli-tool` es opcional y solo se usa para introspección del modelo (preferido: Microsoft Modeling MCP cuando está disponible).
 
 ### 2 MCP Servers (oficiales de Microsoft)
 
 | MCP | Para qué |
 |---|---|
-| `powerbi-modeling` | Conexión local a Power BI Desktop (XMLA). Leer/escribir modelo semántico, ejecutar DAX. |
+| `powerbi-modeling-mcp` | Conexión local a Power BI Desktop (XMLA). Leer/escribir modelo semántico, ejecutar DAX. |
 | `microsoft-learn` | HTTP MCP de Microsoft Learn — docs oficiales de Power BI/DAX/Fabric en contexto de la conversación. |
 
 ---
 
 ## Agentes soportados
 
-| Agente | Skills | MCP config path |
-|---|---|---|
-| **Claude Code** | ✅ `~/.claude/skills/` | `~/.claude.json` |
-| **GitHub Copilot** | ✅ `~/.copilot/skills/` | `~/.copilot/mcp-config.json` |
-| **Codex (OpenAI)** | ✅ `~/.agents/skills/` | `~/.codex/config.toml` |
-| **Gemini CLI** | ✅ `~/.gemini/skills/` | `~/.gemini/settings.json` |
-| **Kilo Code** | ✅ `~/.kilo/skills/` | `~/.kilo/mcp_settings.json` |
+| Agente | Skills | MCP config path | Cómo se usa |
+|---|---|---|---|
+| **Claude Code** | ✅ via `/plugin install` o `~/.claude/skills/` | `~/.claude.json` | Lenguaje natural o slash commands si corrés `super kickoff` |
+| **GitHub Copilot** | ✅ `~/.copilot/skills/` | `~/.copilot/mcp-config.json` | Lenguaje natural |
+| **Codex (OpenAI)** | ✅ `~/.agents/skills/` | `~/.codex/config.toml` | Lenguaje natural |
+| **Gemini CLI** | ✅ `~/.gemini/skills/` | `~/.gemini/settings.json` | Lenguaje natural |
+| **Kilo Code** | ✅ `~/.kilo/skills/` | `~/.kilo/mcp_settings.json` | Lenguaje natural |
 
-`super install` escribe todo esto automáticamente.
+`super install` (npm path) escribe los skills + MCPs automáticamente para todos los agentes detectados.
 
 ### Instalación selectiva
 
@@ -79,12 +106,35 @@ super install --all --yes
 
 ## Actualizar
 
+Dos caminos según cómo instalaste:
+
+### Claude Code (instalado via marketplace)
+
+Dentro de Claude Code:
+
+```
+/plugin update bi-superpowers
+```
+
+Eso trae la última versión publicada en el marketplace. Cero interacción con npm.
+
+### npm (todos los otros agentes)
+
 ```bash
-# Actualizar la CLI a la última versión
-npm install -g @luquimbo/bi-superpowers@latest
+# 1. Actualizá la CLI a la última versión
+super upgrade
 
-# Re-instalar skills y MCPs (sobrescribe los viejos)
-super install --all --yes
+# 2. Refrescá skills + MCPs en tus agentes (no se hace automático, es un paso separado)
+super install --yes
+```
+
+`super upgrade` reinstala el paquete global y al final te imprime un recordatorio con este segundo paso. No lo hace automático a propósito: los usuarios que solo quieren actualizar la CLI (p.ej. desarrollando local) pueden saltarlo.
+
+Equivalente manual si `super upgrade` falla:
+
+```bash
+npm install -g @luquimbo/bi-superpowers@latest
+super install --yes
 ```
 
 ---
@@ -111,7 +161,8 @@ Si ya instalaste los MCPs por alguno de estos caminos, no necesitás correr `sup
 ## Requisitos
 
 - **Node.js ≥ 18**
-- **Power BI Desktop en Windows** (para el MCP `powerbi-modeling` — el MCP `microsoft-learn` funciona en cualquier plataforma)
+- **Power BI Desktop en Windows** (standalone, no Microsoft Store) para workflows locales contra Desktop (`powerbi-modeling-mcp` y `report-design`)
+- **Python 3.10+**, **`pipx`** y **`pbi-cli-tool`** son **opcionales** — solo hacen falta para introspección del modelo via XMLA cuando el Microsoft Modeling MCP no está disponible. Para authoring de visuales, `report-design` usa scripts Node propios y no depende de esa cadena.
 - Al menos uno de: Claude Code, GitHub Copilot, Codex, Gemini CLI, Kilo Code
 
 ---
@@ -119,11 +170,11 @@ Si ya instalaste los MCPs por alguno de estos caminos, no necesitás correr `sup
 ## Comandos
 
 ```bash
-super install          # Instalá skills + MCPs en tus agentes AI
-super kickoff          # Generá el plugin de Claude Code en un proyecto (avanzado)
+super install          # Instalá 4 skills + 2 MCPs en tus agentes AI
+super kickoff          # Generá el plugin local de Claude Code con slash commands
 super recharge         # Regenerá el plugin tras editar skills fuente
 super powers           # Listá skills disponibles
-super upgrade          # Actualizá la CLI a la última versión
+super upgrade          # Actualizá la CLI a la última versión + imprime recordatorio de `super install`
 super help             # Ayuda completa
 ```
 

package/bin/build-plugin.js

@@ -84,7 +84,14 @@ async function main() {
   verifyPluginOutputs(packageDir);
 }
 
-main().catch((error) => {
-  console.error(error.message);
-  process.exit(1);
-});
+// Export for unit testing. The script is also invoked directly during
+// `npm run build:plugin` — we only run main() when it's the entry point,
+// otherwise the test suite would trigger a full build on every import.
+module.exports = { verifyPluginOutputs, main };
+
+if (require.main === module) {
+  main().catch((error) => {
+    console.error(error.message);
+    process.exit(1);
+  });
+}

package/bin/cli.js

@@ -28,6 +28,34 @@ const path = require('path');
 const { execSync } = require('child_process');
 const { loadSkills } = require('./lib/skills');
 
+// Background update check. update-notifier spawns a detached child process
+// that fetches the latest version from npm and caches it; the banner that
+// `notify()` schedules is printed at process exit, so this is zero-latency
+// on the command path. Errors are swallowed silently — the CLI must not
+// break because of a failed update check.
+let _updateNotifierNotify = null;
+try {
+  const updateNotifier = require('update-notifier');
+  const pkg = require('../package.json');
+  const notifier = updateNotifier({
+    pkg,
+    updateCheckInterval: 1000 * 60 * 60 * 24, // 24h
+    shouldNotifyInNpmScript: false,
+  });
+  _updateNotifierNotify = () =>
+    notifier.notify({
+      defer: true,
+      isGlobal: true,
+      message:
+        'Update available {currentVersion} → {latestVersion}\n' +
+        'Run {updateCommand} to update,\n' +
+        'then `super install --yes` to propagate the new skills.',
+    });
+} catch (_) {
+  // update-notifier not available during npm install phase, or network
+  // probe failed — skip silently.
+}
+
 // Optional lib import — may not be available during npm install phase.
 let generators;
 try {
@@ -132,6 +160,14 @@ function main() {
   const args = process.argv.slice(2);
   const command = args[0] || 'help';
 
+  // Schedule the update-notifier banner to print at process exit. Only fire
+  // for "normal" commands — skipping machine-readable and ultra-hot paths so
+  // tools that parse super output don't get surprised by the banner.
+  const SKIP_NOTIFY_COMMANDS = new Set(['version', 'help', 'about', 'info']);
+  if (_updateNotifierNotify && !SKIP_NOTIFY_COMMANDS.has(command)) {
+    _updateNotifierNotify();
+  }
+
   if (commands[command]) {
     commands[command](args.slice(1));
   } else {
@@ -157,7 +193,7 @@ Usage:
   super <command> [options]
 
 Primary commands:
-  install           Install the 2 skills + 2 MCPs across your AI agents
+  install           Install the 4 skills + 2 MCPs across your AI agents
   kickoff [path]    Generate the full Claude Code plugin in a project
   recharge [path]   Regenerate the plugin after editing source skills
   build-desktop     Build the .mcpb extension for Claude Desktop
@@ -558,27 +594,83 @@ function showCompletionMessage(targetDir, tools, skillCount) {
 }
 
 /**
- * `super upgrade` — reinstall the package at the latest version using
- * whichever package manager the user ran npm with.
+ * Detect the package manager the user ran from the npm_config_user_agent.
+ * Returns the pm name ('npm', 'pnpm', 'yarn', 'bun') and the install command
+ * string appropriate for a global install to @latest.
+ *
+ * Exported (via module.exports) so tests can exercise the detection without
+ * spawning a real install.
  */
-function updatePackage() {
-  console.log(`Updating ${PACKAGE_NAME}...\n`);
+function getUpgradeCommand(userAgent) {
+  const agent = userAgent || '';
+  let pm = 'npm';
+  if (agent.includes('pnpm')) pm = 'pnpm';
+  else if (agent.includes('yarn')) pm = 'yarn';
+  else if (agent.includes('bun')) pm = 'bun';
+
+  const cmd =
+    pm === 'yarn'
+      ? `yarn global add ${PACKAGE_NAME}@latest`
+      : `${pm} install -g ${PACKAGE_NAME}@latest`;
+
+  return { pm, cmd };
+}
 
+/**
+ * `super upgrade` — reinstall the package at the latest version.
+ *
+ * After the reinstall, prints a "next steps" block with both update paths
+ * (Claude Code `/plugin update` vs npm-based `super install --yes`) so
+ * users know how to propagate the new skills into their agents. We
+ * intentionally don't auto-chain into `super install`:
+ *   - Claude Code users who installed via the plugin marketplace refresh
+ *     from inside Claude Code (`/plugin update bi-superpowers`), they don't
+ *     need `super install`.
+ *   - npm users want to be able to opt out of re-propagation (e.g. testing
+ *     a version before rolling it across all agents).
+ */
+/**
+ * Wipe ~/.bi-superpowers/update-state.json after a successful upgrade so
+ * the new version starts with a clean cache (no stale snooze TTL, no
+ * cached "latest" from the previous version). Best-effort: any failure
+ * is swallowed — the reset is a convenience, not a correctness gate.
+ *
+ * Separate function + exported so tests can exercise it without spawning
+ * a real `super upgrade`.
+ */
+function resetUpdateCheckStateAfterUpgrade(homeDir) {
   try {
-    const userAgent = process.env.npm_config_user_agent || '';
-    let pm = 'npm';
-    if (userAgent.includes('pnpm')) pm = 'pnpm';
-    else if (userAgent.includes('yarn')) pm = 'yarn';
-    else if (userAgent.includes('bun')) pm = 'bun';
+    const { resetState } = require('./commands/update-check');
+    resetState(path.join(homeDir, '.bi-superpowers'));
+    return true;
+  } catch (_) {
+    return false;
+  }
+}
 
-    const cmd =
-      pm === 'yarn'
-        ? `yarn global add ${PACKAGE_NAME}@latest`
-        : `${pm} install -g ${PACKAGE_NAME}@latest`;
+function updatePackage() {
+  console.log(`Updating ${PACKAGE_NAME}...\n`);
 
+  try {
+    const { cmd } = getUpgradeCommand(process.env.npm_config_user_agent);
     console.log(`Running: ${cmd}\n`);
     execSync(cmd, { stdio: 'inherit' });
-    console.log('\n✓ Update complete.');
+    console.log('\n✓ CLI upgraded.');
+
+    // Clear the update-check cache + snooze state so the newly-installed
+    // version gets a fresh read on next skill invocation.
+    resetUpdateCheckStateAfterUpgrade(require('os').homedir());
+
+    console.log('\nNext step — refresh the skills + MCPs in your agents:\n');
+    console.log('  Claude Code (installed via marketplace):');
+    console.log('    /plugin update bi-superpowers');
+    console.log('');
+    console.log('  Any other install path (npm + super install):');
+    console.log('    super install --yes');
+    console.log('');
+    console.log(
+      'If you only wanted the CLI updated (e.g. developing locally), you can skip the above.'
+    );
   } catch (error) {
     console.error('Update failed:', error.message);
     console.log(`\nTry manually: npm install -g ${PACKAGE_NAME}@latest`);
@@ -710,9 +802,14 @@ module.exports = {
   getSkillFiles,
   loadToolConfig,
   saveToolConfig,
+  getUpgradeCommand,
+  resetUpdateCheckStateAfterUpgrade,
   AI_TOOLS,
   SKILLS_DIR,
   VERSION,
 };
 
-main();
+// Only run main() when invoked as the binary (not when required from tests).
+if (require.main === module) {
+  main();
+}

package/bin/commands/build-desktop.js

@@ -4,7 +4,7 @@
  *
  * Builds a .mcpb extension for Claude Desktop by:
  * 1. Copying desktop-extension/ template to a temp directory
- * 2. Copying all 24 skill sources into temp/skills/
+ * 2. Copying every skill source (flat + folder-based) into temp/skills/
  * 3. Installing dependencies
  * 4. Running `npx @anthropic-ai/mcpb pack` to create the .mcpb file
  * 5. Moving the result to the current directory
@@ -18,6 +18,7 @@ const fs = require('fs');
 const path = require('path');
 const os = require('os');
 const { execSync } = require('child_process');
+const { loadSkills } = require('../lib/skills');
 
 /** Directory where the npm package is installed */
 const PACKAGE_DIR = path.dirname(path.dirname(__dirname));
@@ -25,8 +26,27 @@ const PACKAGE_DIR = path.dirname(path.dirname(__dirname));
 /** Desktop extension template source */
 const DESKTOP_TEMPLATE_DIR = path.join(PACKAGE_DIR, 'desktop-extension');
 
-/** Skill source files directory */
-const SKILLS_SOURCE_DIR = path.join(PACKAGE_DIR, 'src', 'content', 'skills');
+/**
+ * Bundle skills into the Desktop extension's skills/ directory.
+ *
+ * Handles both flat `src/content/skills/<name>.md` and folder-based
+ * `src/content/skills/<name>/SKILL.md` layouts by flattening each skill
+ * to a single `<name>.md` file inside `skillsOutDir`. The Claude Desktop
+ * MCP server loads those flat files as prompts — `references/` and
+ * `scripts/` are distributed through `super install`, not bundled here.
+ *
+ * @param {string} skillsOutDir - Destination directory (created if missing)
+ * @param {string} packageDir - npm package root containing src/content/skills/
+ * @returns {string[]} Names of skills bundled into skillsOutDir
+ */
+function bundleSkills(skillsOutDir, packageDir) {
+  fs.mkdirSync(skillsOutDir, { recursive: true });
+  const skills = loadSkills({ packageDir });
+  for (const skill of skills) {
+    fs.writeFileSync(path.join(skillsOutDir, `${skill.name}.md`), skill.content);
+  }
+  return skills.map((s) => s.name);
+}
 
 /**
  * Build the .mcpb extension for Claude Desktop.
@@ -34,7 +54,7 @@ const SKILLS_SOURCE_DIR = path.join(PACKAGE_DIR, 'src', 'content', 'skills');
  * @param {string[]} args - CLI arguments
  * @param {Object} config - Command config from CLI
  */
-module.exports = function buildDesktop(_args, _config) {
+function buildDesktop(_args, _config) {
   const outputDir = process.cwd();
 
   console.log(`
@@ -50,7 +70,8 @@ BI Agent Superpowers — Build Desktop Extension
   }
 
   // Verify skills exist
-  if (!fs.existsSync(SKILLS_SOURCE_DIR)) {
+  const skillsSourceDir = path.join(PACKAGE_DIR, 'src', 'content', 'skills');
+  if (!fs.existsSync(skillsSourceDir)) {
     console.error(
       'Error: skill sources not found. Try reinstalling: npm install -g @luquimbo/bi-superpowers'
     );
@@ -71,15 +92,10 @@ BI Agent Superpowers — Build Desktop Extension
       }
     }
 
-    // Copy all skill .md files into temp/skills/
-    const skillsOutDir = path.join(tmpDir, 'skills');
-    fs.mkdirSync(skillsOutDir, { recursive: true });
-
-    const skillFiles = fs.readdirSync(SKILLS_SOURCE_DIR).filter((f) => f.endsWith('.md'));
-    for (const file of skillFiles) {
-      fs.copyFileSync(path.join(SKILLS_SOURCE_DIR, file), path.join(skillsOutDir, file));
-    }
-    console.log(`  Bundled ${skillFiles.length} skills`);
+    // Bundle every skill (flat + folder-based) into temp/skills/
+    const bundledNames = bundleSkills(path.join(tmpDir, 'skills'), PACKAGE_DIR);
+    const skillFiles = bundledNames.map((n) => `${n}.md`);
+    console.log(`  Bundled ${bundledNames.length} skills`);
 
     // Patch template placeholder versions with the real package version
     const pkgVersion = require(path.join(PACKAGE_DIR, 'package.json')).version;
@@ -149,7 +165,7 @@ BI Agent Superpowers — Build Desktop Extension
     2. Go to Settings > Extensions > Install Extension
 
   The extension adds ${skillFiles.length} BI skills as MCP prompts.
-  Use the "setup-mcp" prompt for Power BI/Fabric MCP configuration.
+  Use the "setup-mcp" prompt for Power BI Modeling MCP configuration.
 `);
     } else {
       console.log('\n  Build complete. Check current directory for .mcpb file.');
@@ -165,4 +181,7 @@ BI Agent Superpowers — Build Desktop Extension
       // Ignore cleanup errors
     }
   }
-};
+}
+
+module.exports = buildDesktop;
+module.exports.bundleSkills = bundleSkills;

package/bin/commands/diff.js

@@ -12,6 +12,7 @@
 const fs = require('fs');
 const path = require('path');
 const tui = require('../utils/tui');
+const { readSkillDirectory, normalizeSkillName } = require('../lib/skills');
 
 /**
  * Generate a simple diff between two strings
@@ -122,7 +123,12 @@ function compareFiles(skillPath, generatedPath) {
 function getGeneratedPath(skillName, tool, targetDir) {
   switch (tool) {
     case 'claude-code':
-      return path.join(targetDir, '.claude', 'commands', `${skillName}.md`);
+      // The Claude Code plugin generator (bin/lib/generators/claude-plugin.js)
+      // writes slash commands to `<targetDir>/commands/<skill>.md` — NOT
+      // `.claude/commands/...`. An earlier path here pointed at the old
+      // location and silently made `super scan --tool claude-code`
+      // report every skill as "not-generated".
+      return path.join(targetDir, 'commands', `${skillName}.md`);
     case 'cursor':
       // Cursor uses a single file, can't diff per-skill
       return null;
@@ -249,23 +255,27 @@ function diffCommand(args, config) {
     return;
   }
 
-  // Get skill files to compare
+  // Get skill files to compare. Use the shared skill loader so both
+  // flat (`<name>.md`) and folder-based (`<name>/SKILL.md`) skills
+  // are picked up — a previous version filtered with
+  // `f.endsWith('.md')` and silently dropped every folder-based skill.
+  const allSkills = readSkillDirectory(skillsDir);
+  const skillsByName = new Map(allSkills.map((s) => [s.name, s]));
+  const skillNameByPath = new Map(allSkills.map((s) => [s.path, s.name]));
   let skillFiles = [];
 
   if (options.skill) {
-    const skillName = options.skill.replace('.md', '');
-    const skillPath = path.join(skillsDir, `${skillName}.md`);
-    if (fs.existsSync(skillPath)) {
-      skillFiles = [skillPath];
+    // normalizeSkillName accepts `dax`, `dax.md`, `report-design`, or
+    // `report-design/SKILL.md` and returns the canonical skill name.
+    const skill = skillsByName.get(normalizeSkillName(options.skill));
+    if (skill) {
+      skillFiles = [skill.path];
     } else {
       tui.error(`Skill not found: ${options.skill}`);
       process.exit(1);
     }
   } else {
-    skillFiles = fs
-      .readdirSync(skillsDir)
-      .filter((f) => f.endsWith('.md'))
-      .map((f) => path.join(skillsDir, f));
+    skillFiles = allSkills.map((s) => s.path);
   }
 
   if (skillFiles.length === 0) {
@@ -276,9 +286,11 @@ function diffCommand(args, config) {
   tui.info(`Comparing ${skillFiles.length} skill(s) for ${options.tool}...`);
   tui.section('Changes');
 
-  // Compare each skill
+  // Compare each skill. Use the loader-built map to derive the canonical
+  // skill name — for folder-based skills the path is `<folder>/SKILL.md`
+  // and `path.basename(...)` would return `SKILL`, breaking the diff.
   const results = skillFiles.map((skillPath) => {
-    const skillName = path.basename(skillPath, '.md');
+    const skillName = skillNameByPath.get(skillPath) || path.basename(skillPath, '.md');
     const generatedPath = getGeneratedPath(skillName, options.tool, targetDir);
 
     if (!generatedPath) {
@@ -324,4 +336,10 @@ function diffCommand(args, config) {
   }
 }
 
-module.exports = diffCommand;
+// Expose primitives for tests (B3 — replace fake tests with real ones).
+module.exports = Object.assign(diffCommand, {
+  generateDiff,
+  compareFiles,
+  getGeneratedPath,
+  parseArgs,
+});

package/bin/commands/install.js

@@ -470,11 +470,15 @@ async function installCommand(args, config) {
         '\n' +
         chalk.gray('Reiniciá tu agente AI para que tome los MCPs nuevos.') +
         '\n\n' +
-        chalk.gray('Los 2 skills disponibles:') +
+        chalk.gray('Los 4 skills disponibles:') +
         '\n' +
-        chalk.gray('  /project-kickoff  — Analizá tu proyecto BI') +
+        chalk.gray('  /bi-start         — Arrancar una sesión: menú + update + conexión') +
         '\n' +
-        chalk.gray('  /pbi-connect      — Conectá tu agente a Power BI Desktop'),
+        chalk.gray('  /project-kickoff  — Analizá tu proyecto BI (proyecto nuevo)') +
+        '\n' +
+        chalk.gray('  /pbi-connect      — Conectá tu agente a Power BI Desktop') +
+        '\n' +
+        chalk.gray('  /report-design    — Generá reportes PBIR para Power BI Desktop (Windows)'),
       {
         padding: 1,
         margin: { top: 1 },