@liriraid/agentflow-ai 1.0.21 → 1.0.23

package/README.md

@@ -32,7 +32,7 @@ project-workspace/
 |-------|------|----------|-------|
 | **Claude-Orchestrator** | Session coordinator | - | Never implements code directly; delegates to workers |
 | **Codex** | Primary implementation | 1st choice | Structured tasks, tests, docs |
-| **OpenCode** | Analysis + Implementation | 2nd choice | Uses Mistral Medium 3.5 128B for coding |
+| **OpenCode** | Analysis + Implementation | 2nd choice | Uses your configured model |
 | **Claude-Worker** (Backend/Frontend) | Fallback implementation | 3rd choice | Takes over when Codex/OpenCode fail |
 | **Gemini** | Code review/audits | Optional | Disabled by default |
 | **Cursor/Abacus** | Mechanical tasks | Optional | Disabled by default |
@@ -44,7 +44,7 @@ project-workspace/
 
 ### 4. **Smart Task Delegation**
 - **Analysis tasks** → Always assigned to **OpenCode**.
-- **Implementation tasks** → Assigned to **Codex** (1st) → **OpenCode** (2nd, if using Mistral Medium 3.5 128B) → **Claude-Worker** (3rd).
+- **Implementation tasks** → Assigned to **Codex** (1st) → **OpenCode** (2nd) → **Claude-Worker** (3rd).
 - **Fallback chain**: `Codex → OpenCode → Claude-Worker` (automatic).
 
 ### 5. **Persistent Memory & SDD**
@@ -56,14 +56,14 @@ project-workspace/
 
 ### Global CLI (Recommended)
 ```bash
-npm i -g @liriraid/agentflow-ai
+pnpm add -g @liriraid/agentflow-ai
 ```
 
 ### Local Development
 ```bash
 git clone https://github.com/LiriRaid/agentflow-ai.git
 cd agentflow-ai
-npm install
+pnpm install
 ```
 
 ## 🛠️ Quick Start
@@ -169,17 +169,16 @@ orchestrator-my-project/
     "opencode": { "enabled": true, "localConfigDir": ".opencode" }
   },
   "agents": {
-    "Backend": { "cli": "claude", "defaultRepo": "backend", "model": "sonnet" },
-    "Frontend": { "cli": "claude", "defaultRepo": "frontend", "model": "sonnet" },
-    "Codex": { "cli": "codex", "defaultRepo": "backend", "model": "gpt-5.5" },
-    "OpenCode": { "cli": "opencode", "defaultRepo": "frontend", "model": "auto" }
+    "Backend": { "cli": "claude", "defaultRepo": "backend" },
+    "Frontend": { "cli": "claude", "defaultRepo": "frontend" },
+    "Codex": { "cli": "codex", "defaultRepo": "backend" },
+    "OpenCode": { "cli": "opencode", "defaultRepo": "frontend" }
   }
 }
 ```
 
 ### Model Selection
-- Use `"model": "auto"` to let the agent use your default configured model (e.g., Mistral Medium 3.5 128B for OpenCode).
-- Specify a model explicitly (e.g., `"model": "gpt-5.5"`) to override.
+Each agent uses the model you have configured in its own CLI. You can optionally override it by adding `"model": "model-name"` to any agent entry — for Claude this accepts values like `sonnet` or `opus`.
 
 ## 🔄 Workflow Example
 
@@ -201,12 +200,12 @@ orchestrator-my-project/
 
 ## 📊 Supported Agents & Models
 
-| Agent | CLI | Default Model | Implementation? | Notes |
-|-------|-----|----------------|----------------|-------|
-| Backend | `claude` | sonnet | ✅ Yes | Claude-Worker for backend tasks |
-| Frontend | `claude` | sonnet | ✅ Yes | Claude-Worker for frontend tasks |
-| Codex | `codex` | gpt-5.5 | ✅ Yes | Primary implementation |
-| OpenCode | `opencode` | auto | ✅ **Yes** (with Mistral Medium 3.5 128B) | Secondary implementation |
+| Agent | CLI | Model | Implementation? | Notes |
+|-------|-----|-------|----------------|-------|
+| Backend | `claude` | auto | ✅ Yes | Claude-Worker for backend tasks |
+| Frontend | `claude` | auto | ✅ Yes | Claude-Worker for frontend tasks |
+| Codex | `codex` | auto | ✅ Yes | Primary implementation |
+| OpenCode | `opencode` | auto | ✅ Yes | Secondary implementation |
 | Gemini | `gemini` | auto | ❌ No | Audits/reviews only |
 | Cursor | `cursor` | auto | ❌ No | Bulk edits only |
 | Abacus | `abacusai` | auto | ❌ No | Small focused tasks |

package/orchestrator.js

@@ -33,13 +33,11 @@ const CONFIG_TEMPLATE = {
     Backend: {
       cli: "claude",
       defaultRepo: "backend",
-      model: "sonnet",
       instructionsFile: "agents/BACKEND.md",
     },
     Frontend: {
       cli: "claude",
       defaultRepo: "frontend",
-      model: "sonnet",
       instructionsFile: "agents/FRONTEND.md",
     },
     Codex: {

package/package.json

@@ -1,60 +1,63 @@
-{
-  "name": "@liriraid/agentflow-ai",
-  "version": "1.0.21",
-  "description": "Multi-agent workspace orchestrator with TUI. Coordinates AI coding agents over your real frontend and backend projects.",
-  "author": "LiriRaid",
-  "homepage": "https://github.com/LiriRaid/agentflow-ai#readme",
-  "repository": {
-    "type": "git",
-    "url": "git+https://github.com/LiriRaid/agentflow-ai.git"
-  },
-  "bugs": {
-    "url": "https://github.com/LiriRaid/agentflow-ai/issues"
-  },
-  "publishConfig": {
-    "access": "public"
-  },
-  "main": "orchestrator.js",
-  "bin": {
-    "agentflow": "bin/agentflow.mjs"
-  },
-  "files": [
-    "bin",
-    "src",
-    "templates",
-    "orchestrator.js",
-    "LICENSE"
-  ],
-  "scripts": {
-    "start": "node orchestrator.js",
-    "start:paused": "node orchestrator.js --paused",
-    "start:ink": "node src/ink/index.mjs",
-    "start:ink:paused": "node src/ink/index.mjs --paused",
-    "cli:help": "node bin/agentflow.mjs --help"
-  },
-  "keywords": [
-    "ai",
-    "agents",
-    "orchestrator",
-    "agentflow-ai",
-    "claude",
-    "codex",
-    "gemini",
-    "cursor",
-    "opencode",
-    "abacus",
-    "tui",
-    "multi-agent",
-    "coding-assistant",
-    "openspec",
-    "engram",
-    "skills",
-    "agent-config"
-  ],
-  "license": "MIT",
-  "dependencies": {
-    "blessed": "^0.1.81",
-    "ink": "^5.2.1",
-    "react": "^18.3.1"
-  }
-}
+{
+  "name": "@liriraid/agentflow-ai",
+  "version": "1.0.23",
+  "description": "Multi-agent workspace orchestrator with TUI. Coordinates AI coding agents over your real frontend and backend projects.",
+  "author": "LiriRaid",
+  "homepage": "https://github.com/LiriRaid/agentflow-ai#readme",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/LiriRaid/agentflow-ai.git"
+  },
+  "bugs": {
+    "url": "https://github.com/LiriRaid/agentflow-ai/issues"
+  },
+  "publishConfig": {
+    "access": "public"
+  },
+  "main": "orchestrator.js",
+  "bin": {
+    "agentflow": "bin/agentflow.mjs"
+  },
+  "files": [
+    "bin",
+    "src",
+    "templates",
+    "orchestrator.js",
+    "LICENSE"
+  ],
+  "engines": {
+    "node": ">=18.0.0"
+  },
+  "keywords": [
+    "ai",
+    "agents",
+    "orchestrator",
+    "agentflow-ai",
+    "claude",
+    "codex",
+    "gemini",
+    "cursor",
+    "opencode",
+    "abacus",
+    "tui",
+    "multi-agent",
+    "coding-assistant",
+    "openspec",
+    "engram",
+    "skills",
+    "agent-config"
+  ],
+  "license": "MIT",
+  "dependencies": {
+    "blessed": "^0.1.81",
+    "ink": "^5.2.1",
+    "react": "^18.3.1"
+  },
+  "scripts": {
+    "start": "node orchestrator.js",
+    "start:paused": "node orchestrator.js --paused",
+    "start:ink": "node src/ink/index.mjs",
+    "start:ink:paused": "node src/ink/index.mjs --paused",
+    "cli:help": "node bin/agentflow.mjs --help"
+  }
+}

package/src/ink/index.mjs

@@ -100,6 +100,7 @@ let quitRequested = false;
 let resizeTimer = null;
 let isResizing = false;
 let lastColumns = process.stdout?.columns ?? 0;
+let isRendering = false;
 
 function normalizeInlineMessage(message) {
 	return String(message ?? '')
@@ -324,21 +325,24 @@ function ensureEngine() {
 }
 
 function refresh() {
-	if (isResizing) return;
-
-	const snapshot = buildSnapshot();
-	lastRenderedSnapshot = snapshot;
-	ensureStateWatcher();
-	if (!inkApp) {
-		inkApp = render(React.createElement(App, {snapshot, onAction: requestAction}), {
-			exitOnCtrlC: false,
-			patchConsole: false,
-			alternateScreen: true
-		});
-		return;
+	if (isResizing || isRendering) return;
+	isRendering = true;
+	try {
+		const snapshot = buildSnapshot();
+		lastRenderedSnapshot = snapshot;
+		ensureStateWatcher();
+		if (!inkApp) {
+			inkApp = render(React.createElement(App, {snapshot, onAction: requestAction}), {
+				exitOnCtrlC: false,
+				patchConsole: false,
+				alternateScreen: true
+			});
+			return;
+		}
+		inkApp.rerender(React.createElement(App, {snapshot, onAction: requestAction}));
+	} finally {
+		isRendering = false;
 	}
-
-	inkApp.rerender(React.createElement(App, {snapshot, onAction: requestAction}));
 }
 
 function clearTerminal() {
@@ -415,31 +419,25 @@ function ensureStateWatcher() {
 	} catch { stateWatcher = null; }
 }
 
-// Tick independiente del reloj: actualiza activeLabel y timestamp cada 1s
-// sin leer el STATE_FILE, usando la última snapshot cacheada.
+// Tick del reloj: solo actualiza lastRenderedSnapshot en memoria.
+// scheduleRefresh es la única fuente de rerender para evitar renders concurrentes.
 function startClockTick() {
 	clockInterval = setInterval(() => {
-		if (!inkApp || !lastRenderedSnapshot?.startedAt) return;
+		if (!lastRenderedSnapshot?.startedAt) return;
 		const activeSeconds = Math.max(0, Math.round((Date.now() - lastRenderedSnapshot.startedAt) / 1000));
-		const updated = {
+		lastRenderedSnapshot = {
 			...lastRenderedSnapshot,
 			activeLabel: formatDuration(activeSeconds),
 			timestamp: new Date().toLocaleString(getLocale(), {hour12: false})
 		};
-		lastRenderedSnapshot = updated;
-		inkApp.rerender(React.createElement(App, {snapshot: updated, onAction: requestAction}));
 	}, 1000);
 }
 
 function scheduleRefresh() {
-	const engineState = readEngineState();
-	const busy = engineState && Object.values(engineState.agents || {}).some(a => a.status === 'busy');
-	const hasWork = engineState && ((engineState.queue || []).length > 0 || (engineState.inProgress || []).length > 0);
-	const ms = (busy || hasWork) ? 1000 : 4000;
 	refreshTimer = setTimeout(() => {
 		refresh();
 		scheduleRefresh();
-	}, ms);
+	}, 1000);
 }
 
 function mount() {

package/templates/en/AGENT-CONFIG.md

@@ -53,7 +53,7 @@ Default routing should put executable work on Codex or OpenCode first when they
 
 - `claude` is the primary profile.
 - `codex` is the primary implementation profile
-- `opencode` is a support profile that **can implement code when using advanced models** (e.g., Mistral Medium 3.5 128B)
+- `opencode` is a support profile that **can implement code** depending on the model configured in your OpenCode installation
 - Other profiles can exist while remaining disabled by default.
 
 ## Suggested Local Folders
@@ -72,5 +72,5 @@ If both global and local agent configuration exist, the local project config sho
 
 - Claude uses `.claude/skills/` as the main project skill base.
 - Codex and OpenCode can have their own local config even if they do not use the same skill model today.
-- OpenCode can explore, audit, and **implement code when using advanced models** like Mistral Medium 3.5 128B.
+- OpenCode can explore, audit, and **implement code** — capabilities depend on your configured model.
 - The design should allow richer local layers for additional agents later.

package/templates/en/ORCHESTRATOR.md

@@ -122,7 +122,7 @@ del .away-mode
 The TUI handles fallback automatically following this chain:
 
 ```
-Codex fails → OpenCode (with Mistral Medium 3.5 128B) → Frontend (frontend repo) or Backend (backend repo)
+Codex fails → OpenCode → Frontend (frontend repo) or Backend (backend repo)
 ```
 Codex fails  →  Frontend (frontend repo) or Backend (backend repo) directly
 ```
@@ -149,14 +149,13 @@ Default agent summary:
 | Backend | claude | Backend code through Claude-Worker |
 | Frontend | claude | Broad frontend work through Claude-Worker |
 | Codex | codex | **First choice for implementation**; structured implementation, tests, docs, narrow frontend support |
-| OpenCode | opencode | **Second choice for implementation** (with Mistral Medium 3.5 128B); exploration, audits, structured reports |
+| OpenCode | opencode | **Second choice for implementation**; exploration, audits, structured reports |
 | Gemini | gemini | Optional audits and reviews only when explicitly enabled |
 | Cursor | cursor | Optional mechanical bulk edits only when explicitly enabled |
 | Abacus | abacusai | Optional small focused tasks only when explicitly enabled |
 
 **Notes about OpenCode:**
-- When OpenCode uses **Mistral Medium 3.5 128B or equivalent models**, it can implement code.
-- If the model is not suitable for implementation, OpenCode will only perform analysis and report as `blocked`.
+- OpenCode can perform both analysis and implementation — capabilities depend on the model you have configured in your OpenCode installation.
 
 ## How To Assign Work
 
@@ -184,9 +183,9 @@ Rules:
 
 Routing preferences:
 
-1. Use OpenCode for exploration, audits, and **implementation (with Mistral Medium 3.5 128B or advanced models)**.
+1. Use OpenCode for exploration, audits, and **implementation**.
 2. Use Codex as the **primary implementation agent** when the spec is clear.
-3. Use OpenCode as the **secondary implementation agent** (when using advanced models).
+3. Use OpenCode as the **secondary implementation agent**.
 4. Keep Claude-Worker available as automatic fallback for Codex/OpenCode and for overflow tasks.
 5. For frontend, use Codex for narrow tasks and Frontend/Claude-Worker for broad UI work.
 6. Do not assign all tasks to Claude just because Claude is the orchestrator.

package/templates/en/README.md

@@ -28,12 +28,12 @@ project-workspace/
 - Agents work on the real project files via absolute paths configured in `orchestrator.config.json`.
 
 ### 2. **Multi-Agent Coordination**
-| Agent | CLI | Default Model | Implementation? | Notes |
-|-------|-----|----------------|----------------|-------|
+| Agent | CLI | Model | Implementation? | Notes |
+|-------|-----|-------|----------------|-------|
 | **Claude-Orchestrator** | - | - | ❌ No | Session coordinator only |
-| **Codex** | `codex` | gpt-5.5 | ✅ Yes | Primary implementation |
-| **OpenCode** | `opencode` | auto | ✅ **Yes** (with Mistral Medium 3.5 128B) | Secondary implementation |
-| **Claude-Worker** (Backend/Frontend) | `claude` | sonnet | ✅ Yes | Fallback implementation |
+| **Codex** | `codex` | auto | ✅ Yes | Primary implementation |
+| **OpenCode** | `opencode` | auto | ✅ Yes | Secondary implementation |
+| **Claude-Worker** (Backend/Frontend) | `claude` | auto | ✅ Yes | Fallback implementation |
 | **Gemini** | `gemini` | auto | ❌ No | Audits/reviews only |
 | **Cursor** | `cursor` | auto | ❌ No | Bulk edits only |
 | **Abacus** | `abacusai` | auto | ❌ No | Small focused tasks |
@@ -45,7 +45,7 @@ project-workspace/
 
 ### 4. **Smart Task Delegation**
 - **Analysis tasks** → Always assigned to **OpenCode**.
-- **Implementation tasks** → Assigned to **Codex** (1st) → **OpenCode** (2nd, if using Mistral Medium 3.5 128B) → **Claude-Worker** (3rd).
+- **Implementation tasks** → Assigned to **Codex** (1st) → **OpenCode** (2nd) → **Claude-Worker** (3rd).
 - **Fallback chain**: `Codex → OpenCode → Claude-Worker` (automatic).
 
 ### 5. **Persistent Memory & SDD**
@@ -57,14 +57,14 @@ project-workspace/
 
 ### Global CLI (Recommended)
 ```bash
-npm i -g @liriraid/agentflow-ai
+pnpm add -g @liriraid/agentflow-ai
 ```
 
 ### Local Development
 ```bash
 git clone https://github.com/LiriRaid/agentflow-ai.git
 cd agentflow-ai
-npm install
+pnpm install
 ```
 
 ## 🛠️ Quick Start
@@ -170,17 +170,16 @@ orchestrator-my-project/
     "opencode": { "enabled": true, "localConfigDir": ".opencode" }
   },
   "agents": {
-    "Backend": { "cli": "claude", "defaultRepo": "backend", "model": "sonnet" },
-    "Frontend": { "cli": "claude", "defaultRepo": "frontend", "model": "sonnet" },
-    "Codex": { "cli": "codex", "defaultRepo": "backend", "model": "gpt-5.5" },
-    "OpenCode": { "cli": "opencode", "defaultRepo": "frontend", "model": "auto" }
+    "Backend": { "cli": "claude", "defaultRepo": "backend" },
+    "Frontend": { "cli": "claude", "defaultRepo": "frontend" },
+    "Codex": { "cli": "codex", "defaultRepo": "backend" },
+    "OpenCode": { "cli": "opencode", "defaultRepo": "frontend" }
   }
 }
 ```
 
 ### Model Selection
-- Use `"model": "auto"` to let the agent use your default configured model (e.g., Mistral Medium 3.5 128B for OpenCode).
-- Specify a model explicitly (e.g., `"model": "gpt-5.5"`) to override.
+Each agent uses the model you have configured in its own CLI. You can optionally add `"model": "model-name"` to any agent entry to override — for Claude this accepts values like `sonnet` or `opus`.
 
 ## 🔄 Workflow Example
 
@@ -195,19 +194,19 @@ orchestrator-my-project/
    - Reads OpenCode's report.
    - Creates `TASK-002` (Codex): `"Implement JWT auth"` (depends on TASK-001).
 5. **Codex/OpenCode**:
-   - Implements the feature (Codex first, OpenCode second if using Mistral Medium 3.5 128B).
+   - Implements the feature (Codex first, OpenCode second).
    - Reports completion in `progress/PROGRESS-*.md`.
 6. **TUI**:
    - Shows real-time updates (task status, agent activity, costs).
 
 ## 📊 Supported Agents & Models
 
-| Agent | CLI | Default Model | Implementation? | Notes |
-|-------|-----|----------------|----------------|-------|
-| Backend | `claude` | sonnet | ✅ Yes | Claude-Worker for backend tasks |
-| Frontend | `claude` | sonnet | ✅ Yes | Claude-Worker for frontend tasks |
-| Codex | `codex` | gpt-5.5 | ✅ Yes | Primary implementation |
-| OpenCode | `opencode` | auto | ✅ **Yes** (with Mistral Medium 3.5 128B) | Secondary implementation |
+| Agent | CLI | Model | Implementation? | Notes |
+|-------|-----|-------|----------------|-------|
+| Backend | `claude` | auto | ✅ Yes | Claude-Worker for backend tasks |
+| Frontend | `claude` | auto | ✅ Yes | Claude-Worker for frontend tasks |
+| Codex | `codex` | auto | ✅ Yes | Primary implementation |
+| OpenCode | `opencode` | auto | ✅ Yes | Secondary implementation |
 | Gemini | `gemini` | auto | ❌ No | Audits/reviews only |
 | Cursor | `cursor` | auto | ❌ No | Bulk edits only |
 | Abacus | `abacusai` | auto | ❌ No | Small focused tasks |

package/templates/en/agents/OPENCODE.md

@@ -2,12 +2,8 @@
 
 ## Role
 
-OpenCode is a **multi-purpose agent** that can perform **analysis, exploration AND CODE IMPLEMENTATION**. 
-When using powerful models like **Mistral Medium 3.5 128B**, OpenCode is capable of:
-- Analyzing existing code
-- Implementing new features
-- Fixing complex bugs
-- Refactoring code
+OpenCode is a **multi-purpose agent** capable of **analysis, exploration, and code implementation**.
+Its capabilities depend on the model you have configured in your OpenCode installation.
 
 ## Scope
 
@@ -19,7 +15,7 @@ When using powerful models like **Mistral Medium 3.5 128B**, OpenCode is capable
 - Structured Markdown reports
 - Identifying dead code, missing dependencies, inconsistencies
 
-### Implementation (With advanced models like Mistral Medium 3.5 128B)
+### Implementation
 - Implementing new features
 - Modifying project files
 - Writing new tests
@@ -36,20 +32,9 @@ When using powerful models like **Mistral Medium 3.5 128B**, OpenCode is capable
 ## Assignment Priority
 
 - **First choice for implementation**: Codex (when available)
-- **Second choice for implementation**: OpenCode (with Mistral Medium 3.5 128B or equivalent models)
+- **Second choice for implementation**: OpenCode
 - **Third choice**: Claude-Worker (Backend/Frontend)
 
-## Model-Specific Behavior
-
-### With analysis models (e.g., small models or low-context):
-- **Analysis only**: Do not implement code
-- Report in TASK_REPORT: `status: blocked`, `issues: "Model not suitable for implementation — reassign to Codex or Claude-Worker"`
-
-### With implementation models (e.g., Mistral Medium 3.5 128B, GPT-4, etc.):
-- **You CAN implement code** when the task is clearly defined
-- Ensure that any required prior analysis is already complete
-- Follow the same quality rules as Codex
-
 ## Completion Report (REQUIRED)
 
 ### For analysis tasks:
@@ -75,12 +60,3 @@ summary: 1-3 sentences describing the changes made
 issues: problems or "none"
 TASK_REPORT_END
 ```
-TASK_REPORT
-status: completed | failed | blocked
-files_modified: none
-files_created: none
-files_deleted: none
-summary: 1-3 sentences describing findings
-issues: problems or "none"
-TASK_REPORT_END
-```

package/templates/en/docs/usage.md

@@ -5,7 +5,7 @@
 ### 1. Install the CLI
 
 ```bash
-npm i -g @liriraid/agentflow-ai
+pnpm add -g @liriraid/agentflow-ai
 ```
 
 ### 2. Create a sibling orchestrator workspace

package/templates/en/orchestrator.config.json

@@ -50,49 +50,42 @@
       "cli": "claude",
       "profile": "claude",
       "defaultRepo": "backend",
-      "model": "sonnet",
       "instructionsFile": "agents/BACKEND.md"
     },
     "Frontend": {
       "cli": "claude",
       "profile": "claude",
       "defaultRepo": "frontend",
-      "model": "sonnet",
       "instructionsFile": "agents/FRONTEND.md"
     },
     "Codex": {
       "cli": "codex",
       "profile": "codex",
       "defaultRepo": "frontend",
-      "model": "gpt-5.5",
       "instructionsFile": "agents/CODEX.md"
     },
     "Gemini": {
       "cli": "gemini",
       "profile": "gemini",
       "defaultRepo": "frontend",
-      "model": "auto",
       "instructionsFile": "agents/GEMINI.md"
     },
     "OpenCode": {
       "cli": "opencode",
       "profile": "opencode",
       "defaultRepo": "frontend",
-      "model": "auto",
       "instructionsFile": "agents/OPENCODE.md"
     },
     "Cursor": {
       "cli": "cursor",
       "profile": "cursor",
       "defaultRepo": "backend",
-      "model": "auto",
       "instructionsFile": "agents/CURSOR.md"
     },
     "Abacus": {
       "cli": "abacusai",
       "profile": "abacusai",
       "defaultRepo": "backend",
-      "model": "auto",
       "instructionsFile": "agents/ABACUS.md"
     }
   }

package/templates/es/AGENT-CONFIG.md

@@ -59,7 +59,7 @@ Para este proyecto reusable:
 
 - `claude` es el profile principal
 - `codex` es el profile de implementación primaria
-- `opencode` es el profile de apoyo que **puede implementar código cuando usa modelos avanzados** (ej: Mistral Medium 3.5 128B)
+- `opencode` es el profile de apoyo que **puede implementar código** dependiendo del modelo configurado en tu instalación de OpenCode
 - otros profiles pueden existir, aunque no estén habilitados por defecto
 
 ## Directorios locales sugeridos
@@ -80,5 +80,5 @@ Si existe configuración global del agente en el home del usuario y también una
 
 - `Claude` usa `.claude/skills/` como base principal del proyecto
 - `Codex` y `OpenCode` pueden tener configuración local propia aunque hoy no usen el mismo modelo de skills
-- `OpenCode` no es solo auditor: puede explorar, auditar e **implementar código cuando use modelos avanzados** como Mistral Medium 3.5 128B
+- `OpenCode` no es solo auditor: puede explorar, auditar e **implementar código** — las capacidades dependen del modelo configurado en tu instalación de OpenCode
 - el diseño debe permitir que mañana también tengan una capa local más rica

package/templates/es/ORCHESTRATOR.md

@@ -135,7 +135,7 @@ del .away-mode
 La TUI gestiona el fallback automáticamente siguiendo esta cadena:
 
 ```
-Codex falla → OpenCode (con Mistral Medium 3.5 128B) → Frontend (repo FE) o Backend (repo BE)
+Codex falla → OpenCode → Frontend (repo FE) o Backend (repo BE)
 ```
 Codex falla  →  Frontend (repo FE) o Backend (repo BE) directamente
 ```
@@ -166,17 +166,16 @@ Revisa `orchestrator.config.json` → `agents`. Cada entrada tiene:
 **Agentes por defecto en esta plantilla:**
 | Nombre | CLI | Mejor para |
 |--------|-----|------------|
-| Backend | claude (sonnet) | Código server-side: controllers, models, migrations y tests |
-| Frontend | claude (sonnet) | Código UI: componentes, páginas y estilos |
+| Backend | claude | Código server-side: controllers, models, migrations y tests |
+| Frontend | claude | Código UI: componentes, páginas y estilos |
 | Codex | codex | **Primera opción para implementación**; docs, migraciones y tareas estructuradas con spec clara |
-| OpenCode | opencode | **Segunda opción para implementación** (con Mistral Medium 3.5 128B); exploración, auditorías y reportes estructurados |
+| OpenCode | opencode | **Segunda opción para implementación**; exploración, auditorías y reportes estructurados |
 | Gemini | gemini | Auditorías, code review; suele sufrir con `node_modules` muy grandes |
 | Cursor | cursor | Tareas mecánicas de alto volumen: find-and-replace y cleanup |
 | Abacus | abacusai | Tareas pequeñas y enfocadas, con alcance bien acotado |
 
 **Notas sobre OpenCode:**
-- Cuando OpenCode usa **Mistral Medium 3.5 128B o modelos equivalentes**, puede implementar código.
-- Si el modelo no es apto para implementación, OpenCode solo hará análisis y reportará como `blocked`.
+- OpenCode puede realizar tanto análisis como implementación — las capacidades dependen del modelo que tengas configurado en tu instalación de OpenCode.
 
 ## Cómo asignar trabajo
 
@@ -197,7 +196,7 @@ Revisa `orchestrator.config.json` → `agents`. Cada entrada tiene:
 4. (Opcional) Para un brief muy detallado, crea `briefs/TASK-NNN-BRIEF.md`; también se inyecta.
 5. Dependencias: agrega `> after:TASK-NNN` al final de la descripción para bloquear la tarea.
 6. **La TUI inicia automáticamente** - NO necesitas presionar R ni S. La TUI detecta nuevas tasks y las lanza.
-7. **Codex es la primera opción para implementación; OpenCode (con Mistral Medium 3.5 128B) es la segunda opción.** Claude-Worker es el fallback automático de Codex/OpenCode y también toma trabajo cuando hay más tareas que agentes disponibles.
+7. **Codex es la primera opción para implementación; OpenCode es la segunda opción.** Claude-Worker es el fallback automático de Codex/OpenCode y también toma trabajo cuando hay más tareas que agentes disponibles.
 8. **REGLA CRÍTICA SOBRE ANÁLISIS:** Si OpenCode ya analizó algo y escribió su reporte en `INBOX.md` o `progress/PROGRESS-OpenCode.md`, **TÚ (Claude-Orquestador) NO DEBES VOLVER A ANALIZAR EL MISMO CÓDIGO**. Usa el reporte existente para crear tareas de implementación. Si necesitas más detalles, pide a OpenCode que haga un análisis adicional con una nueva TASK, pero **NUNCA lo hagas tú directamente**.
 9. Distribución según cantidad de TASKs independientes:
    - **1 tarea de análisis**: OpenCode.
@@ -217,7 +216,7 @@ Revisa `orchestrator.config.json` → `agents`. Cada entrada tiene:
 6. Al terminar la sesión, escribe un `handoffs/HANDOFF-<fecha>.md` resumiendo qué se hizo y qué sigue.
 7. **Por defecto solo usa Claude, Codex y OpenCode**. No uses Gemini, Cursor ni Abacus salvo instrucción explícita del usuario.
 8. Si el usuario activa **Modo Ausencia**, revisa progreso cada 5 minutos y reasigna nuevas TASKs razonables dentro del alcance actual sin esperar confirmación intermedia.
-9. La TUI gestiona el fallback automáticamente: Codex falla → OpenCode (si usa Mistral Medium 3.5 128B) → Claude-Worker (Frontend/Backend según repo). Solo intervén manualmente si la tarea queda marcada `failed`.
+9. La TUI gestiona el fallback automáticamente: Codex falla → OpenCode → Claude-Worker (Frontend/Backend según repo). Solo intervén manualmente si la tarea queda marcada `failed`.
 10. Usa Engram para guardar decisiones, hallazgos, bugs y resúmenes de sesión; no dependas solo del contexto corto de la conversación.
 11. **VERIFICACIÓN OBLIGATORIA:** Antes de crear cualquier TASK de implementación, **LEE Y CONFIRMA QUE:**
    - Existe un reporte de OpenCode en `INBOX.md` o `progress/PROGRESS-OpenCode.md` para el análisis solicitado.

package/templates/es/PROJECT.md

@@ -44,7 +44,7 @@ Esto evita ensuciar el repo del producto con:
 La instalación recomendada para usuarios finales es:
 
 ```bash
-npm i -g @liriraid/agentflow-ai
+pnpm add -g @liriraid/agentflow-ai
 ```
 
 Después, por cada proyecto real:
@@ -61,7 +61,7 @@ La variante con `npx` sigue siendo válida:
 npx @liriraid/agentflow-ai-ai init-workspace C:/code/mi-proyecto
 ```
 
-No se recomienda `npm install @liriraid/agentflow-ai-ai  dentro del repo del producto, porque eso lo vuelve una dependencia local del proyecto en vez de una herramienta global del entorno.
+No se recomienda `pnpm add @liriraid/agentflow-ai` dentro del repo del producto, porque eso lo vuelve una dependencia local del proyecto en vez de una herramienta global del entorno.
 
 ## Regla de permisos
 

package/templates/es/README.md

@@ -34,7 +34,7 @@ workspace-proyecto/
 |--------|-----|----------|-------|
 | **Claude-Orquestador** | Coordinador de sesión | - | Nunca implementa código directamente; delega a workers |
 | **Codex** | Implementación primaria | 1ra opción | Tareas estructuradas, tests, docs |
-| **OpenCode** | Análisis + Implementación | 2da opción | Usa Mistral Medium 3.5 128B para código |
+| **OpenCode** | Análisis + Implementación | 2da opción | Usa el modelo que tengas configurado |
 | **Claude-Worker** (Backend/Frontend) | Fallback | 3ra opción | Toma el relevo si Codex/OpenCode fallan |
 | **Gemini** | Revisión/auditoría | Opcional | Deshabilitado por defecto |
 | **Cursor/Abacus** | Tareas mecánicas | Opcional | Deshabilitado por defecto |
@@ -46,7 +46,7 @@ workspace-proyecto/
 
 ### 4. **Delegación Inteligente de Tareas**
 - **Tareas de análisis** → Siempre asignadas a **OpenCode**.
-- **Tareas de implementación** → Asignadas a **Codex** (1ra) → **OpenCode** (2da, si usa Mistral Medium 3.5 128B) → **Claude-Worker** (3ra).
+- **Tareas de implementación** → Asignadas a **Codex** (1ra) → **OpenCode** (2da) → **Claude-Worker** (3ra).
 - **Cadena de fallback**: `Codex → OpenCode → Claude-Worker` (automático).
 
 ### 5. **Memoria Persistente y SDD**
@@ -58,14 +58,14 @@ workspace-proyecto/
 
 ### CLI Global (Recomendado)
 ```bash
-npm i -g @liriraid/agentflow-ai
+pnpm add -g @liriraid/agentflow-ai
 ```
 
 ### Desarrollo Local
 ```bash
 git clone https://github.com/LiriRaid/agentflow-ai.git
 cd agentflow-ai
-npm install
+pnpm install
 ```
 
 ## 🛠️ Inicio Rápido
@@ -171,17 +171,16 @@ orchestrator-mi-proyecto/
     "opencode": { "enabled": true, "localConfigDir": ".opencode" }
   },
   "agents": {
-    "Backend": { "cli": "claude", "defaultRepo": "backend", "model": "sonnet" },
-    "Frontend": { "cli": "claude", "defaultRepo": "frontend", "model": "sonnet" },
-    "Codex": { "cli": "codex", "defaultRepo": "backend", "model": "gpt-5.5" },
-    "OpenCode": { "cli": "opencode", "defaultRepo": "frontend", "model": "auto" }
+    "Backend": { "cli": "claude", "defaultRepo": "backend" },
+    "Frontend": { "cli": "claude", "defaultRepo": "frontend" },
+    "Codex": { "cli": "codex", "defaultRepo": "backend" },
+    "OpenCode": { "cli": "opencode", "defaultRepo": "frontend" }
   }
 }
 ```
 
 ### Selección de Modelo
-- Usa `"model": "auto"` para que el agente use el modelo configurado por defecto en tu sistema (ej: Mistral Medium 3.5 128B para OpenCode).
-- Especifica un modelo explícitamente (ej: `"model": "gpt-5.5"`) para sobrescribir.
+Cada agente usa el modelo que tengas configurado en su propio CLI. Podés agregar opcionalmente `"model": "nombre-del-modelo"` en la entrada de cualquier agente para sobrescribir — para Claude acepta valores como `sonnet` u `opus`.
 
 ## 🔄 Ejemplo de Flujo de Trabajo
 
@@ -203,12 +202,12 @@ orchestrator-mi-proyecto/
 
 ## 📊 Agentes Soportados y Modelos
 
-| Agente | CLI | Modelo por Defecto | ¿Implementa? | Notas |
-|--------|-----|---------------------|--------------|-------|
-| Backend | `claude` | sonnet | ✅ Sí | Claude-Worker para tareas de backend |
-| Frontend | `claude` | sonnet | ✅ Sí | Claude-Worker para tareas de frontend |
-| Codex | `codex` | gpt-5.5 | ✅ Sí | Implementación primaria |
-| OpenCode | `opencode` | auto | ✅ **Sí** (con Mistral Medium 3.5 128B) | Implementación secundaria |
+| Agente | CLI | Modelo | ¿Implementa? | Notas |
+|--------|-----|--------|--------------|-------|
+| Backend | `claude` | auto | ✅ Sí | Claude-Worker para tareas de backend |
+| Frontend | `claude` | auto | ✅ Sí | Claude-Worker para tareas de frontend |
+| Codex | `codex` | auto | ✅ Sí | Implementación primaria |
+| OpenCode | `opencode` | auto | ✅ Sí | Implementación secundaria |
 | Gemini | `gemini` | auto | ❌ No | Solo auditorías/revisiones |
 | Cursor | `cursor` | auto | ❌ No | Solo ediciones masivas |
 | Abacus | `abacusai` | auto | ❌ No | Solo tareas pequeñas y enfocadas |
@@ -290,7 +289,7 @@ pero **ese no es el flujo recomendado** para este proyecto.
 Instala el CLI una sola vez:
 
 ```bash
-npm i -g @liriraid/agentflow-ai
+pnpm add -g @liriraid/agentflow-ai
 ```
 
 Luego, para cada proyecto real, crea un workspace sibling del orquestador:
@@ -330,7 +329,7 @@ Si vas a modificar el orquestador mismo:
 ```bash
 git clone https://github.com/LiriRaid/agentflow.git
 cd agentflow
-npm install
+pnpm install
 ```
 
 Este repo local es la **fuente reusable** que tú modificas para agregar herramientas, cambiar el flujo o extender el sistema.
@@ -363,7 +362,7 @@ También crea carpetas runtime:
 ### 1. Instalar el CLI globalmente
 
 ```bash
-npm i -g @liriraid/agentflow-ai
+pnpm add -g @liriraid/agentflow-ai
 ```
 
 ### 2. Crear el workspace del orquestador
@@ -691,14 +690,12 @@ Eso asegura:
       "cli": "claude",
       "profile": "claude",
       "defaultRepo": "backend",
-      "model": "sonnet",
       "instructionsFile": "agents/BACKEND.md"
     },
     "Codex": {
       "cli": "codex",
       "profile": "codex",
       "defaultRepo": "backend",
-      "model": "gpt-5.5",
       "instructionsFile": "agents/CODEX.md"
     },
     "OpenCode": {

package/templates/es/agents/OPENCODE.md

@@ -2,12 +2,8 @@
 
 ## Rol
 
-OpenCode es un agente **multipropósito** que puede realizar **análisis, exploración E IMPLEMENTACIÓN de código**. 
-Cuando usa modelos potentes como **Mistral Medium 3.5 128B**, OpenCode es capaz de:
-- Analizar código existente
-- Implementar nuevas funcionalidades
-- Resolver bugs complejos
-- Refactorizar código
+OpenCode es un agente **multipropósito** capaz de realizar **análisis, exploración e implementación de código**.
+Sus capacidades dependen del modelo que tengas configurado en tu instalación de OpenCode.
 
 ## Alcance
 
@@ -19,7 +15,7 @@ Cuando usa modelos potentes como **Mistral Medium 3.5 128B**, OpenCode es capaz
 - Reportes estructurados en Markdown
 - Identificación de residuos, dependencias faltantes, inconsistencias
 
-### Implementación (Con modelos avanzados como Mistral Medium 3.5 128B)
+### Implementación
 - Implementar nuevas features
 - Modificar archivos del proyecto
 - Escribir tests nuevos
@@ -36,20 +32,9 @@ Cuando usa modelos potentes como **Mistral Medium 3.5 128B**, OpenCode es capaz
 ## Prioridad de Asignación
 
 - **Primera opción para implementación**: Codex (cuando esté disponible)
-- **Segunda opción para implementación**: OpenCode (con Mistral Medium 3.5 128B o modelos equivalentes)
+- **Segunda opción para implementación**: OpenCode
 - **Tercera opción**: Claude-Worker (Backend/Frontend)
 
-## Comportamiento según el Modelo
-
-### Con modelos de análisis (ej: modelos pequeños o de bajo contexto):
-- **Solo análisis**: No implementes código
-- Reporta en TASK_REPORT: `status: blocked`, `issues: "Modelo no apto para implementación — reasignar a Codex o Claude-Worker"`
-
-### Con modelos de implementación (ej: Mistral Medium 3.5 128B, GPT-4, etc.):
-- **Puedes implementar código** cuando la tarea esté claramente definida
-- Asegúrate de que el análisis previo (si era necesario) ya esté completo
-- Sigue las mismas reglas de calidad que Codex
-
 ## Reporte de finalización (OBLIGATORIO)
 
 ### Para tareas de análisis:

package/templates/es/docs/usage.md

@@ -5,7 +5,7 @@
 ### 1. Instalar el CLI globalmente
 
 ```bash
-npm i -g @liriraid/agentflow-ai
+pnpm add -g @liriraid/agentflow-ai
 ```
 
 ### 2. Crear un workspace hermano para un proyecto

package/templates/es/orchestrator.config.json

@@ -50,49 +50,42 @@
       "cli": "claude",
       "profile": "claude",
       "defaultRepo": "backend",
-      "model": "sonnet",
       "instructionsFile": "agents/BACKEND.md"
     },
     "Frontend": {
       "cli": "claude",
       "profile": "claude",
       "defaultRepo": "frontend",
-      "model": "sonnet",
       "instructionsFile": "agents/FRONTEND.md"
     },
     "Codex": {
       "cli": "codex",
       "profile": "codex",
       "defaultRepo": "frontend",
-      "model": "gpt-5.5",
       "instructionsFile": "agents/CODEX.md"
     },
-"Gemini": {
+    "Gemini": {
       "cli": "gemini",
       "profile": "gemini",
       "defaultRepo": "frontend",
-      "model": "auto",
       "instructionsFile": "agents/GEMINI.md"
     },
     "OpenCode": {
       "cli": "opencode",
       "profile": "opencode",
       "defaultRepo": "frontend",
-      "model": "auto",
       "instructionsFile": "agents/OPENCODE.md"
     },
     "Cursor": {
       "cli": "cursor",
       "profile": "cursor",
       "defaultRepo": "backend",
-      "model": "auto",
       "instructionsFile": "agents/CURSOR.md"
     },
     "Abacus": {
       "cli": "abacusai",
       "profile": "abacusai",
       "defaultRepo": "backend",
-      "model": "auto",
       "instructionsFile": "agents/ABACUS.md"
     }
   }