refactor agents system_prompts

minerva.ipynb

@@ -28,7 +28,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 72,
+   "execution_count": 103,
    "metadata": {},
    "outputs": [
     {
@@ -40,17 +40,15 @@
     }
    ],
    "source": [
-    "%pip install -q autogen-agentchat~=0.2 pillow pytesseract"
+    "%pip install -q autogen-agentchat~=0.2 pillow pytesseract pyyaml"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 73,
+   "execution_count": null,
    "metadata": {},
    "outputs": [],
    "source": [
-    "from typing_extensions import Annotated\n",
-    "\n",
     "import autogen\n",
     "\n",
     "from IPython.display import Image as IPImage\n",
@@ -59,7 +57,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 74,
+   "execution_count": 106,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -81,6 +79,18 @@
     "}"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": 107,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import yaml\n",
+    "\n",
+    "with open('config/agents.yaml', 'r') as file:\n",
+    "    config = yaml.safe_load(file)"
+   ]
+  },
   {
    "cell_type": "markdown",
    "metadata": {},
@@ -90,50 +100,33 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 99,
+   "execution_count": 108,
    "metadata": {},
    "outputs": [],
    "source": [
     "def create_agents():\n",
-    "    \"\"\"Create and initialize the specialized agents.\"\"\"\n",
     "\n",
     "    ocr_agent = autogen.AssistantAgent(\n",
     "        name=\"OCR_Specialist\",\n",
-    "        system_message=\"\"\"You are an OCR specialist. Your role is to:\n",
-    "        1. Extract text from an image path using pytesseract\n",
-    "        2. Clean and format the extracted text for further analysis\"\"\",\n",
+    "        system_message=config['ocr_agent']['assignment'],\n",
     "        llm_config=llm_config\n",
     "    )\n",
     "   \n",
     "    content_agent = autogen.AssistantAgent(\n",
     "        name=\"Content_Analyst\",\n",
-    "        system_message=\"\"\"You are a content analysis specialist. Your role is to:\n",
-    "        1. Analyze text for common scam patterns\n",
-    "        2. Identify and analyze URLs, phone numbers, or other contact information\n",
-    "        3. Identify urgency indicators, threats, or pressure tactics\n",
-    "        5. Check for inconsistencies in messaging\n",
-    "        6. Evaluate legitimacy of any claims or offers\"\"\",\n",
+    "        system_message=config['content_agent']['assignment'],\n",
     "        llm_config=llm_config\n",
     "    )\n",
     "\n",
     "    decision_agent = autogen.AssistantAgent(\n",
     "        name=\"Decision_Maker\",\n",
-    "        system_message=\"\"\"You are the final decision maker. Your role is to:\n",
-    "        1. Coordinate with other agents to gather all necessary information\n",
-    "        2. Make a final determination on scam probability\n",
-    "        3. Provide detailed explanation of the decision\n",
-    "        4. End your explanation with the label as 'TASK_COMPLETE' when done\"\"\",\n",
+    "        system_message=config['decision_agent']['assignment'],\n",
     "        llm_config=llm_config\n",
     "    )\n",
     "\n",
     "    summary_agent = autogen.AssistantAgent(\n",
     "        name=\"Summary_Agent\",\n",
-    "        system_message=\"\"\"You are a communication specialist who creates clear, concise summaries of technical analyses. Your role is to:\n",
-    "        1. Synthesize the findings of a scam assessment into user-friendly language\n",
-    "        2. Highlight the most important points that users need to know\n",
-    "        3. Provide actionable recommendations\n",
-    "        4. Use clear, non-technical language while maintaining accuracy\n",
-    "        5. Format information in a way that's easy to read and understand\"\"\",\n",
+    "        system_message=config['summary_agent']['assignment'],\n",
     "        llm_config=llm_config\n",
     "    )\n",
     "\n",
@@ -169,7 +162,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 100,
+   "execution_count": 109,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -203,7 +196,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 101,
+   "execution_count": 110,
    "metadata": {},
    "outputs": [
     {
@@ -233,7 +226,7 @@
       "\u001b[0m\n",
       "\u001b[33mOCR_Specialist\u001b[0m (to chat_manager):\n",
       "\n",
-      "\u001b[32m***** Suggested tool call (call_ta6tNtpskpnHXvaWicWmYHi8): ocr *****\u001b[0m\n",
+      "\u001b[32m***** Suggested tool call (call_u8NqTuVd9I0eMW0FEyuggyxJ): ocr *****\u001b[0m\n",
       "Arguments: \n",
       "{\"image_path\":\"./samples/02.giftcard.message.scam.png\"}\n",
       "\u001b[32m********************************************************************\u001b[0m\n",
@@ -246,7 +239,7 @@
       ">>>>>>>> EXECUTING FUNCTION ocr...\u001b[0m\n",
       "\u001b[33muser_proxy\u001b[0m (to chat_manager):\n",
       "\n",
-      "\u001b[32m***** Response from calling tool (call_ta6tNtpskpnHXvaWicWmYHi8) *****\u001b[0m\n",
+      "\u001b[32m***** Response from calling tool (call_u8NqTuVd9I0eMW0FEyuggyxJ) *****\u001b[0m\n",
       "Congratulations!\n",
       "You've won a $1,000\n",
       "Walmart gift card. Go\n",
@@ -266,35 +259,32 @@
       "\n",
       "---\n",
       "\n",
-      "**Congratulations!**  \n",
-      "You've won a **$1,000**  \n",
-      "Walmart gift card. Go  \n",
-      "[http://bit.ly/123456](http://bit.ly/123456)  \n",
-      "tp claim now.\n",
+      "**Congratulations!  \n",
+      "You've won a $1,000 Walmart gift card. Go  \n",
+      "http://bit.ly/123456  \n",
+      "to claim now.**\n",
       "\n",
       "---\n",
       "\n",
-      "### Cleaning and Formatting the Extracted Text:\n",
+      "### Analysis of the Messaging and Claims:\n",
       "\n",
-      "1. **Remove unnecessary characters:** The text appears to have an odd break (\"tp claim now.\") which might be due to OCR error.\n",
-      "   \n",
-      "2. **Correct the message structure:** Creating a more readable format.\n",
+      "1. **Claim of Winning a Gift Card:** \n",
+      "   - The message states that the recipient has won a $1,000 gift card which is a common tactic used in scams to entice users with unrealistic prizes.\n",
       "\n",
-      "---\n",
+      "2. **Call-to-Action (CTA):**\n",
+      "   - The instruction to \"**Go to http://bit.ly/123456 to claim now**\" encourages immediate action, which is typically observed in fraudulent messages. This urgency can lead individuals to make hasty decisions.\n",
       "\n",
-      "**Final Cleaned Text:**\n",
+      "3. **Use of a URL Shortener:**\n",
+      "   - The link provided is a shortened URL (bit.ly). Scammers often use such links to obscure the destination, making it harder for users to identify if the site is safe.\n",
       "\n",
-      "```\n",
-      "Congratulations! \n",
-      "You've won a $1,000 Walmart gift card.\n",
-      "Go to http://bit.ly/123456 to claim now.\n",
-      "```\n",
+      "4. **Lack of Official Branding:** \n",
+      "   - Although Walmart is mentioned, there is no official branding or logo associated with this message, which further indicates it may not be legitimate.\n",
       "\n",
-      "### Next Steps:\n",
-      "1. **Evaluate the messaging and claims** based on typical characteristics of scams. \n",
-      "2. **Synthesize analysis** and provide a final determination on the legitimacy of the message. \n",
+      "5. **Common Scam Characteristics:** \n",
+      "   - The format and language (e.g., \"Congratulations!\" and \"You've won\") align with typical scam messages aimed at eliciting personal information or financial details.\n",
       "\n",
-      "Would you like me to proceed with evaluating the messaging and claims?\n",
+      "### Final Determination:\n",
+      "This message is likely part of a scam. It claims an unrealistic prize, employs urgency, uses a URL shortener, and lacks official branding. It is advisable to avoid clicking on such links or providing any personal information.\n",
       "\n",
       "--------------------------------------------------------------------------------\n",
       "\u001b[32m\n",
@@ -302,30 +292,26 @@
       "\u001b[0m\n",
       "\u001b[33mContent_Analyst\u001b[0m (to chat_manager):\n",
       "\n",
-      "### Evaluation of Messaging and Claims:\n",
+      "### Final Analysis & Determination:\n",
       "\n",
-      "1. **Common Scam Patterns:**\n",
-      "   - **Too Good to Be True:** The claim of winning a $1,000 gift card is a classic scam tactic. Scammers often offer substantial rewards to entice victims.\n",
-      "   - **Generic Greeting:** The use of \"Congratulations!\" without personalizing the message is common in mass scam messages.\n",
+      "**1. Common Scam Patterns:**\n",
+      "   - The message claims that the recipient has won a large monetary prize (a $1,000 Walmart gift card), which is a typical scam tactic designed to lure individuals with the promise of free money.\n",
+      "   - The format features exclamatory language and congratulatory tones, often used to evoke excitement and urgency.\n",
       "\n",
-      "2. **Identification of URLs:**\n",
-      "   - The URL provided (`http://bit.ly/123456`) is a shortened URL. Scammers often use URL shorteners to obscure the actual destination, which is a red flag. It is advisable to check where this URL would direct a user since it could lead to phishing sites or malicious downloads.\n",
+      "**2. Contact Information Evaluation:**\n",
+      "   - The provided URL (http://bit.ly/123456) is a shortened link, which is often used in scams to mask the actual destination website. Clicking such links can expose users to malicious websites designed to steal personal or financial information.\n",
       "\n",
-      "3. **Urgency Indicators, Threats, or Pressure Tactics:**\n",
-      "   - Phrases like \"Go to claim now\" indicate a sense of urgency, pressuring the recipient to act quickly without thinking critically about the legitimacy of the offer.\n",
+      "**3. Urgency Indicators:**\n",
+      "   - The phrase \"to claim now\" creates a sense of urgency, pushing the recipient to act quickly without careful consideration. This tactic is commonly employed in scams to disrupt rational decision-making.\n",
       "\n",
-      "4. **Inconsistencies in Messaging:**\n",
-      "   - The abrupt ending with \"tp claim now.\" suggests a potential OCR error or deliberate attempt to confuse the recipient, further indicating a lack of professionalism associated with scams.\n",
+      "**4. Inconsistencies in Messaging:**\n",
+      "   - There is no official branding from Walmart or any legitimate verification in the message, indicating a lack of credibility. Legitimate giveaways by established companies would typically include confirmed branding and often have more detailed processes.\n",
       "\n",
-      "5. **Legitimacy of Claims or Offers:**\n",
-      "   - Large corporations like Walmart do not typically give out gift cards in such arbitrary fashions. Any real promotions would likely require personalization and would not come through unsolicited messages.\n",
+      "**5. Claims and Offers Evaluation:**\n",
+      "   - The claim of winning a $1,000 gift card is highly questionable and likely fraudulent. Such large sums are seldom offered without legitimate contests or promotions. \n",
       "\n",
-      "### Final Determination:\n",
-      "\n",
-      "Based on the analysis of common scam patterns, the use of a shortened URL, urgency indicators, potential errors in the message, and the legitimacy of the claims, it is very likely that this message is a scam. It poses a risk to individuals who might click on the link without verifying the legitimacy of the offer. \n",
-      "\n",
-      "### Recommendation:\n",
-      "Do not engage with the message or click on the link. It is advisable to report such scams to appropriate authorities to help prevent others from falling victim to similar tactics.\n",
+      "### Conclusion:\n",
+      "Based on all observations, this message is highly likely to be a scam. Recipients should disregard the message, avoid clicking the link, and refrain from providing any personal information. It exemplifies the classic traits of phishing scams.\n",
       "\n",
       "--------------------------------------------------------------------------------\n",
       "\u001b[32m\n",
@@ -333,35 +319,32 @@
       "\u001b[0m\n",
       "\u001b[33mDecision_Maker\u001b[0m (to chat_manager):\n",
       "\n",
-      "### Final Decision on Scam Probability\n",
-      "\n",
-      "After gathering and synthesizing information from both the OCR extraction and the content evaluation, I conclude that the probability of this message being a scam is very high.\n",
-      "\n",
-      "#### Detailed Explanation of the Decision:\n",
+      "Based on the analyses provided from both the OCR extraction and content evaluation, I have reached a final determination regarding the probability that this message is a scam.\n",
       "\n",
-      "1. **Content of the Message:**\n",
-      "   - The message claims that the recipient has \"won\" a significant prize, a $1,000 Walmart gift card. This type of claim is a traditional tactic used by scammers to lure individuals into providing personal information.\n",
+      "### Final Determination:\n",
+      "The message purporting to inform the recipient of a $1,000 Walmart gift card win is unequivocally a scam.\n",
       "\n",
-      "2. **Use of a Shortened URL:**\n",
-      "   - The inclusion of a modified URL (`http://bit.ly/123456`) is a major red flag. Scammers tend to use shortened links to hide the destination, leading individuals to malicious websites that can steal information or install malware.\n",
+      "### Detailed Explanation:\n",
       "\n",
-      "3. **Urgency and Pressure:**\n",
-      "   - The message creates a sense of urgency with phrases like \"Go to claim now,\" encouraging immediate action. This tactic is a psychological manipulation commonly seen in scams, meant to prevent the recipient from thinking critically or researching legitimacy.\n",
+      "1. **Classic Scam Traits:**\n",
+      "   - The claim of winning a prize, especially of such an amount, is a hallmark of scams. Organizations do not randomly award gift cards without a proper contest or registration process.\n",
       "\n",
-      "4. **Nature of the Offer:**\n",
-      "   - Offers of large monetary gifts or benefits without any prior engagement or purchase are usually inauthentic. Recognized companies such as Walmart typically have established promotional methods and would not distribute gift cards in this manner.\n",
+      "2. **Urgent Call to Action:**\n",
+      "   - The phrase \"to claim now\" creates a sense of urgency that is typically designed to prevent the target from thinking critically about the offer or engaging in due diligence.\n",
       "\n",
-      "5. **Inconsistencies in Messaging:**\n",
-      "   - The message has grammatical inconsistencies (e.g., “tp” instead of “to”), which can imply poor crafting and possible scams. Legitimate communications from companies are typically well-written and free of such errors.\n",
+      "3. **URL Shortener Usage:**\n",
+      "   - The use of a shortened URL (bit.ly) is another red flag, as it obscures the real link and can lead to harmful sites. Scammers frequently employ such tactics to hide their true intent.\n",
       "\n",
-      "### Final Determination:\n",
+      "4. **Absence of Credibility:**\n",
+      "   - The message lacks official branding and does not direct the recipient to a legitimate platform associated with Walmart. A credible business would provide clear identification and often more detailed instructions for any contests or promotions.\n",
       "\n",
-      "Given the above factors, I assess the probability of this message being a scam to be exceptionally high. It employs multiple red flags characteristic of phishing and scam communications, suggesting that recipients should avoid engaging further and should report it to the relevant authorities.\n",
+      "5. **Emotional Manipulation:**\n",
+      "   - The exclamatory and congratulatory tone is crafted to elicit excitement, preying on emotions to trigger impulsive responses. This strategy is prevalent in scams aiming to elicit personal data or financial information through deceptive means.\n",
       "\n",
       "### Conclusion:\n",
-      "Please refrain from interacting with this message and keep an eye out to identify any attempts to impersonate trusted companies.\n",
+      "All elements analyzed confirm that this is a highly probable scam intended to mislead recipients into providing sensitive information or engaging in fraudulent activity. The message exhibits the quintessential characteristics of phishing attempts, and I recommend immediate disregard for it.\n",
       "\n",
-      "**TASK_COMPLETE**\n",
+      "TASK_COMPLETE\n",
       "\n",
       "--------------------------------------------------------------------------------\n",
       "\u001b[32m\n",
@@ -380,65 +363,58 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 102,
+   "execution_count": 111,
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "{'content': '### Final Decision on Scam Probability\\n'\n",
-      "            '\\n'\n",
-      "            'After gathering and synthesizing information from both the OCR '\n",
-      "            'extraction and the content evaluation, I conclude that the '\n",
-      "            'probability of this message being a scam is very high.\\n'\n",
+      "{'content': 'Based on the analyses provided from both the OCR extraction and '\n",
+      "            'content evaluation, I have reached a final determination '\n",
+      "            'regarding the probability that this message is a scam.\\n'\n",
       "            '\\n'\n",
-      "            '#### Detailed Explanation of the Decision:\\n'\n",
-      "            '\\n'\n",
-      "            '1. **Content of the Message:**\\n'\n",
-      "            '   - The message claims that the recipient has \"won\" a '\n",
-      "            'significant prize, a $1,000 Walmart gift card. This type of claim '\n",
-      "            'is a traditional tactic used by scammers to lure individuals into '\n",
-      "            'providing personal information.\\n'\n",
+      "            '### Final Determination:\\n'\n",
+      "            'The message purporting to inform the recipient of a $1,000 '\n",
+      "            'Walmart gift card win is unequivocally a scam.\\n'\n",
       "            '\\n'\n",
-      "            '2. **Use of a Shortened URL:**\\n'\n",
-      "            '   - The inclusion of a modified URL (`http://bit.ly/123456`) is '\n",
-      "            'a major red flag. Scammers tend to use shortened links to hide '\n",
-      "            'the destination, leading individuals to malicious websites that '\n",
-      "            'can steal information or install malware.\\n'\n",
+      "            '### Detailed Explanation:\\n'\n",
       "            '\\n'\n",
-      "            '3. **Urgency and Pressure:**\\n'\n",
-      "            '   - The message creates a sense of urgency with phrases like \"Go '\n",
-      "            'to claim now,\" encouraging immediate action. This tactic is a '\n",
-      "            'psychological manipulation commonly seen in scams, meant to '\n",
-      "            'prevent the recipient from thinking critically or researching '\n",
-      "            'legitimacy.\\n'\n",
+      "            '1. **Classic Scam Traits:**\\n'\n",
+      "            '   - The claim of winning a prize, especially of such an amount, '\n",
+      "            'is a hallmark of scams. Organizations do not randomly award gift '\n",
+      "            'cards without a proper contest or registration process.\\n'\n",
       "            '\\n'\n",
-      "            '4. **Nature of the Offer:**\\n'\n",
-      "            '   - Offers of large monetary gifts or benefits without any prior '\n",
-      "            'engagement or purchase are usually inauthentic. Recognized '\n",
-      "            'companies such as Walmart typically have established promotional '\n",
-      "            'methods and would not distribute gift cards in this manner.\\n'\n",
+      "            '2. **Urgent Call to Action:**\\n'\n",
+      "            '   - The phrase \"to claim now\" creates a sense of urgency that is '\n",
+      "            'typically designed to prevent the target from thinking critically '\n",
+      "            'about the offer or engaging in due diligence.\\n'\n",
       "            '\\n'\n",
-      "            '5. **Inconsistencies in Messaging:**\\n'\n",
-      "            '   - The message has grammatical inconsistencies (e.g., “tp” '\n",
-      "            'instead of “to”), which can imply poor crafting and possible '\n",
-      "            'scams. Legitimate communications from companies are typically '\n",
-      "            'well-written and free of such errors.\\n'\n",
+      "            '3. **URL Shortener Usage:**\\n'\n",
+      "            '   - The use of a shortened URL (bit.ly) is another red flag, as '\n",
+      "            'it obscures the real link and can lead to harmful sites. Scammers '\n",
+      "            'frequently employ such tactics to hide their true intent.\\n'\n",
       "            '\\n'\n",
-      "            '### Final Determination:\\n'\n",
+      "            '4. **Absence of Credibility:**\\n'\n",
+      "            '   - The message lacks official branding and does not direct the '\n",
+      "            'recipient to a legitimate platform associated with Walmart. A '\n",
+      "            'credible business would provide clear identification and often '\n",
+      "            'more detailed instructions for any contests or promotions.\\n'\n",
       "            '\\n'\n",
-      "            'Given the above factors, I assess the probability of this message '\n",
-      "            'being a scam to be exceptionally high. It employs multiple red '\n",
-      "            'flags characteristic of phishing and scam communications, '\n",
-      "            'suggesting that recipients should avoid engaging further and '\n",
-      "            'should report it to the relevant authorities.\\n'\n",
+      "            '5. **Emotional Manipulation:**\\n'\n",
+      "            '   - The exclamatory and congratulatory tone is crafted to elicit '\n",
+      "            'excitement, preying on emotions to trigger impulsive responses. '\n",
+      "            'This strategy is prevalent in scams aiming to elicit personal '\n",
+      "            'data or financial information through deceptive means.\\n'\n",
       "            '\\n'\n",
       "            '### Conclusion:\\n'\n",
-      "            'Please refrain from interacting with this message and keep an eye '\n",
-      "            'out to identify any attempts to impersonate trusted companies.\\n'\n",
+      "            'All elements analyzed confirm that this is a highly probable scam '\n",
+      "            'intended to mislead recipients into providing sensitive '\n",
+      "            'information or engaging in fraudulent activity. The message '\n",
+      "            'exhibits the quintessential characteristics of phishing attempts, '\n",
+      "            'and I recommend immediate disregard for it.\\n'\n",
       "            '\\n'\n",
-      "            '**TASK_COMPLETE**',\n",
+      "            'TASK_COMPLETE',\n",
       " 'name': 'Decision_Maker',\n",
       " 'role': 'user'}\n"
      ]
@@ -452,7 +428,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 98,
+   "execution_count": 112,
    "metadata": {},
    "outputs": [],
    "source": [