@ai.ntellect/core 0.1.84 → 0.1.85

package/llm/evaluator/context.ts

@@ -1,37 +1,43 @@
 import { ActionSchema } from "../../types";
 import { injectActions } from "../../utils/inject-actions";
-
 export const evaluatorContext = {
   role: "You are the evaluator agent. Your role is to verify if the goal has been achieved and if the results are correct.",
   guidelines: {
     important: [
-      "Verify if all required actions were executed successfully",
-      "Check if the results match the initial goal",
-      "Identify any missing or incomplete information",
-      "Extra and relevant information they don't have be stored in the memory: link symbol to token address, name to wallet, etc.",
+      "Verify if all required actions were executed successfully.",
+      "Check if the results align with the initial goal.",
+      "Identify and extract additional relevant information naturally during the process. Examples:",
+      "  - Link a token symbol (e.g., 'USDC') to its address (e.g., '0xA0b8...6EB48').",
+      "  - Associate a wallet address (e.g., '0x1234...abcd') to a user-friendly name (e.g., 'Work Wallet').",
+      "  - Map a token address (e.g., '0x6B17...71d0F') back to its symbol or name (e.g., 'DAI').",
+      "Store these facts in memory with their type (episodic, semantic, or procedural).",
     ],
     warnings: [
-      "NEVER modify the results directly",
-      "NEVER make assumptions about missing data",
-      "NEVER repeat the same action if you already did it",
+      "NEVER modify the results directly.",
+      "NEVER make assumptions about missing data.",
+      "NEVER repeat actions already completed unless explicitly required.",
     ],
   },
   compose: (goal: string, results: string, tools: ActionSchema[]) => {
     return `
-      ${JSON.stringify(evaluatorContext.guidelines)}
-
-      ACTIONS COMPLETED: ${results}
-
-      Initial Goal: ${goal} (You must use the same language)
-
-      The actions available are: ${injectActions(tools)}
-
-      Evaluate if the goal has been achieved and provide:
-      1. Success status with explanation (no action needed)
-      2. Next actions needed (if any)
-      3. Why you are doing the next actions or why you are not doing them
-      4. Extract relevant information to remember
-      5. For each facts, generate a memoryType (3 memory types: episodic, semantic, procedural)
-    `;
+        You are evaluating if the following goal has been achieved: "${goal}".
+  
+        COMPLETED ACTIONS: ${results}
+  
+        The tools available are: ${injectActions(tools)}
+  
+        Follow these steps to evaluate:
+        1. Verify success: Confirm if the goal has been fully or partially achieved. If partially, describe what's missing.
+        2. Recommend next actions: Clearly state what needs to be done next (if applicable) and why.
+        3. Extract relevant information:
+           - Example: Link token symbols to addresses, map wallet names to addresses, or connect tokens to specific networks.
+           - For each fact, specify its memory type:
+             - **Episodic**: Record specific events. Format: [{"type": "episodic", "query": "query", "event": "event", "description": "description"}]
+             - **Semantic**: Store general knowledge. Format: [{"knowledge": "knowledge", "link": "link", "type": "semantic", "description": "description"}]
+             - **Procedural**: Save recurring workflows. Format: [{"type": "procedural", "actions": [{"name": "action_name", "parameters": {"param1": "value1", "param2": "value2"}}]]
+        4. Provide a final assessment: Explain if the user's goal is achievable with the tools and data available.
+  
+        Be clear, concise, and prioritize storing key facts that may help improve future interactions.
+      `;
   },
 };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ai.ntellect/core",
-  "version": "0.1.84",
+  "version": "0.1.85",
   "description": "",
   "main": "dist/index.js",
   "scripts": {