node-red-contrib-ai-agent 0.3.0 → 0.4.0

package/README.md

@@ -126,10 +126,21 @@ Creates an HTTP request tool that can be used by the AI Agent to make external A
 - **Headers**: JSON object of headers to include in the request
 - **Body**: Content to send in the request body
 
+Example: `https://api.example.com/users/${input.userId}`
+
 **Template Variables:**
 You can use template variables in the URL, headers, and body to reference properties from the input object that the AI provides when calling the tool.
 
-Example: `https://api.example.com/users/${input.userId}`
+### AI Tool Approval
+Creates an approval tool that can be used by the AI Agent to request human intervention. When called, the agent's execution pauses until a human provides a response.
+
+**Properties:**
+- **Name**: Display name for the node
+- **Tool Name**: Name of the tool (used by the AI to identify the tool)
+- **Description**: Description of what the tool does (e.g., "Request approval for payments")
+
+**Output 2 (Approval Request):** Sends a message with `msg.payload` containing the human question and `msg.approvalId` to be used for the response.
+
 
 ## Example: Basic Usage
 

package/orchestrator/orchestrator.js

@@ -91,7 +91,9 @@ Return a JSON object with a "tasks" array. Each task should have:
 - "input": detailed instruction for the agent
 - "status": "pending"
 - "priority": a number (1-10, default 5)
-- "dependsOn": an array of IDs of tasks that must be completed BEFORE this task can start (empty array if none)`;
+- "dependsOn": an array of IDs of tasks that must be completed BEFORE this task can start (empty array if none)
+
+Note: You can use "human_approval" as a task type if you need a human to verify something before proceeding.`;
 
         if (strategy === 'advanced') {
             prompt += `\n\nThink about parallel execution. Group related tasks and identify bottlenecks. Ensure dependencies are logical.`;
@@ -148,7 +150,8 @@ Last Task ${isError ? 'Error' : 'Result'}: ${JSON.stringify(isError ? msg.error
 Evaluate the progress. 
 1. If the last task failed, propose a recovery strategy (retry, alternative task, or fail the goal).
 2. If the goal is achieved, set status to "completed".
-3. Otherwise, continue execution. You may refine the plan by adding, removing, or modifying tasks.
+3. If you need more information or approval from a human, add a task with type "human_approval".
+4. Otherwise, continue execution. You may refine the plan by adding, removing, or modifying tasks.
 
 Return a JSON object:
 {

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "node-red-contrib-ai-agent",
-    "version": "0.3.0",
+    "version": "0.4.0",
     "description": "AI Agent for Node-RED",
     "repository": {
         "type": "git",
@@ -62,7 +62,8 @@
             "ai-tool-http": "./tool-http/ai-tool-http.js",
             "ai-memory-file": "./memory-file/memory-file.js",
             "ai-memory-inmem": "./memory-inmem/memory-inmem.js",
-            "ai-orchestrator": "./orchestrator/orchestrator.js"
+            "ai-orchestrator": "./orchestrator/orchestrator.js",
+            "ai-tool-approval": "./tool-approval/ai-tool-approval.js"
         }
     }
 }