inject prompt if jira tool is detected

Author: abraham2512

bugzooka/analysis/prompts.py

@@ -51,3 +51,20 @@
     ),
     "assistant": "Here is a context-aware analysis of the most relevant failures:",
 }
+
+# Jira tool prompt - used when Jira MCP tools are available
+JIRA_TOOL_PROMPT = {
+    "system_suffix": (
+        "\n\nIMPORTANT: You have access to JIRA search tools. After analyzing the error, "
+        "ALWAYS search for related issues in JIRA using the search_jira_issues tool with the OCPBUGS project. "
+        "Extract key error terms, component names, or operators from the log summary to search for similar issues. "
+        "Include the top 3 most relevant JIRA issues in your final response under a 'Related JIRA Issues' section."
+    ),
+    "user_suffix": (
+        "\n- Related JIRA issues (search using search_jira_issues tool with OCPBUGS project "
+        "and include the top 3 most relevant issues)"
+    ),
+    "assistant_suffix": (
+        "\n\n**Related JIRA Issues:**\n- <Top 3 relevant issues from JIRA search>"
+    ),
+}

bugzooka/integrations/gemini_client.py

@@ -3,7 +3,6 @@
 import ssl
 import httpx
 import json
-import asyncio
 
 from openai import OpenAI
 from langchain_core.utils.function_calling import convert_to_openai_tool
@@ -13,6 +12,7 @@
     INFERENCE_MAX_TOOL_ITERATIONS,
 )
 from bugzooka.integrations.inference import InferenceAPIUnavailableError
+from bugzooka.analysis.prompts import JIRA_TOOL_PROMPT
 
 
 logger = logging.getLogger(__name__)
@@ -127,10 +127,10 @@ async def execute_tool_call(tool_name, tool_args, available_tools):
         logger.info("Executing tool: %s with args: %s", tool_name, tool_args)
 
         # Check if the tool is async (has coroutine attribute or ainvoke method)
-        if hasattr(tool, 'coroutine') and tool.coroutine:
+        if hasattr(tool, "coroutine") and tool.coroutine:
             # MCP tools have a coroutine attribute
             result = await tool.ainvoke(tool_args)
-        elif hasattr(tool, 'ainvoke'):
+        elif hasattr(tool, "ainvoke"):
             # Some tools have ainvoke method
             result = await tool.ainvoke(tool_args)
         else:
@@ -151,7 +151,7 @@ async def analyze_log_with_gemini(
     error_summary: str,
     model="gemini-2.0-flash",
     tools=None,
-    max_iterations=None
+    max_iterations=None,
 ):
     """
     Analyzes log summaries using Gemini API with product-specific prompts and optional tool calling.
@@ -171,26 +171,48 @@ async def analyze_log_with_gemini(
         gemini_client = GeminiClient()
 
         prompt_config = product_config["prompt"][product]
+
+        # Check if Jira MCP tools are available and inject Jira prompt
+        has_jira = tools and any(
+            hasattr(t, "name") and t.name == "search_jira_issues" for t in tools
+        )
+        if has_jira:
+            logger.info("Jira MCP tools detected - injecting Jira prompt")
+            system_prompt = prompt_config["system"] + JIRA_TOOL_PROMPT["system_suffix"]
+            user_suffix = JIRA_TOOL_PROMPT["user_suffix"]
+            assistant_suffix = JIRA_TOOL_PROMPT["assistant_suffix"]
+        else:
+            system_prompt = prompt_config["system"]
+            user_suffix = ""
+            assistant_suffix = ""
+
         try:
-            formatted_content = prompt_config["user"].format(
-                error_summary=error_summary
+            formatted_content = (
+                prompt_config["user"].format(error_summary=error_summary) + user_suffix
             )
         except KeyError:
-            formatted_content = prompt_config["user"].format(summary=error_summary)
+            formatted_content = (
+                prompt_config["user"].format(summary=error_summary) + user_suffix
+            )
 
         messages = [
-            {"role": "system", "content": prompt_config["system"]},
+            {"role": "system", "content": system_prompt},
             {"role": "user", "content": formatted_content},
-            {"role": "assistant", "content": prompt_config["assistant"]},
+            {
+                "role": "assistant",
+                "content": prompt_config["assistant"] + assistant_suffix,
+            },
         ]
 
         # Convert LangChain tools to OpenAI format if provided
         openai_tools = None
         if tools:
             openai_tools = convert_langchain_tools_to_openai_format(tools)
-            logger.info("Enabled %d tools for Gemini: %s",
-                       len(openai_tools),
-                       [t["function"]["name"] for t in openai_tools])
+            logger.info(
+                "Enabled %d tools for Gemini: %s",
+                len(openai_tools),
+                [t["function"]["name"] for t in openai_tools],
+            )
 
         # Tool calling loop - iterate until we get a final answer or hit max iterations
         iteration = 0
@@ -214,7 +236,7 @@ async def analyze_log_with_gemini(
             response_message = response.choices[0].message
 
             # Check if Gemini wants to call tools
-            tool_calls = getattr(response_message, 'tool_calls', None)
+            tool_calls = getattr(response_message, "tool_calls", None)
 
             if not tool_calls:
                 # No tool calls - we have the final answer
@@ -229,21 +251,23 @@ async def analyze_log_with_gemini(
             logger.info("Gemini requested %d tool call(s)", len(tool_calls))
 
             # Add the assistant's message with tool calls to conversation
-            messages.append({
-                "role": "assistant",
-                "content": response_message.content or "",
-                "tool_calls": [
-                    {
-                        "id": tc.id,
-                        "type": "function",
-                        "function": {
-                            "name": tc.function.name,
-                            "arguments": tc.function.arguments
+            messages.append(
+                {
+                    "role": "assistant",
+                    "content": response_message.content or "",
+                    "tool_calls": [
+                        {
+                            "id": tc.id,
+                            "type": "function",
+                            "function": {
+                                "name": tc.function.name,
+                                "arguments": tc.function.arguments,
+                            },
                         }
-                    }
-                    for tc in tool_calls
-                ]
-            })
+                        for tc in tool_calls
+                    ],
+                }
+            )
 
             # Execute each tool call and add results to messages
             for tool_call in tool_calls:
@@ -256,25 +280,25 @@ async def analyze_log_with_gemini(
                 else:
                     # Execute the tool (await since it's now async)
                     function_result = await execute_tool_call(
-                        function_name,
-                        function_args,
-                        tools
+                        function_name, function_args, tools
                     )
 
                 # Add tool result to messages
-                messages.append({
-                    "role": "tool",
-                    "tool_call_id": tool_call.id,
-                    "name": function_name,
-                    "content": function_result
-                })
+                messages.append(
+                    {
+                        "role": "tool",
+                        "tool_call_id": tool_call.id,
+                        "name": function_name,
+                        "content": function_result,
+                    }
+                )
 
             # Continue loop to let Gemini process tool results
 
         # If we hit max iterations without a final answer
         logger.warning(
             "Reached maximum tool calling iterations (%d) without final answer",
-            max_iterations
+            max_iterations,
         )
         return "Analysis incomplete: Maximum tool calling iterations reached. Please try again with a simpler query."
 

prompts.json

@@ -1,8 +1,8 @@
 {
   "OPENSHIFT_PROMPT": {
-    "system": "You are an expert in OpenShift, Kubernetes, and cloud infrastructure. Your task is to analyze logs and summaries related to OpenShift environments. Given a log summary, identify the root cause, potential fixes, and affected components. Be as consise as possible (under 5000 characters), but precise and avoid generic troubleshooting steps. Prioritize OpenShift-specific debugging techniques. Keep in mind that the cluster is ephemeral and is destroyed after the build is complete, but all relevant logs and metrics are available. Use markdown formatting for the output with only one level of bullet points, do not use bold text except for the headers.\n\nIMPORTANT: You have access to JIRA search tools. After analyzing the error, ALWAYS search for related issues in JIRA using the search_jira_issues tool with the OCPBUGS project. Extract key error terms, component names, or operators from the log summary to search for similar issues. Include the top 3 most relevant JIRA issues in your final response under a 'Related JIRA Issues' section.",
-    "user": "Here is the log summary from an OpenShift environment:\n\n{summary}\n\nBased on this summary, provide a structured breakdown of:\n- The OpenShift component likely affected (e.g., etcd, kube-apiserver, ingress, SDN, Machine API)\n- The probable root cause\n- Steps to verify the issue further\n- Suggested resolution, including OpenShift-specific commands or configurations\n- Related JIRA issues (search using search_jira_issues tool with OCPBUGS project and include the top 3 most relevant issues)",
-    "assistant": "**Affected Component:** <Identified component>\n\n**Probable Root Cause:** <Describe why this issue might be occurring>\n\n**Verification Steps:**\n- <Step 1>\n- <Step 2>\n- <Step 3>\n\n**Suggested Resolution:**\n- <OpenShift CLI commands>\n- <Relevant OpenShift configurations>\n\n**Related JIRA Issues:**\n- <Top 3 relevant issues from JIRA search>"
+    "system": "You are an expert in OpenShift, Kubernetes, and cloud infrastructure. Your task is to analyze logs and summaries related to OpenShift environments. Given a log summary, identify the root cause, potential fixes, and affected components. Be as consise as possible (under 5000 characters), but precise and avoid generic troubleshooting steps. Prioritize OpenShift-specific debugging techniques. Keep in mind that the cluster is ephemeral and is destroyed after the build is complete, but all relevant logs and metrics are available. Use markdown formatting for the output with only one level of bullet points, do not use bold text except for the headers.",
+    "user": "Here is the log summary from an OpenShift environment:\n\n{summary}\n\nBased on this summary, provide a structured breakdown of:\n- The OpenShift component likely affected (e.g., etcd, kube-apiserver, ingress, SDN, Machine API)\n- The probable root cause\n- Steps to verify the issue further\n- Suggested resolution, including OpenShift-specific commands or configurations",
+    "assistant": "**Affected Component:** <Identified component>\n\n**Probable Root Cause:** <Describe why this issue might be occurring>\n\n**Verification Steps:**\n- <Step 1>\n- <Step 2>\n- <Step 3>\n\n**Suggested Resolution:**\n- <OpenShift CLI commands>\n- <Relevant OpenShift configurations>"
   },
   "ANSIBLE_PROMPT": {
     "system": "You are an expert in Ansible automation, playbook debugging, and infrastructure as code (IaC). Your task is to analyze log summaries related to Ansible execution, playbook failures, and task errors. Given a log summary, identify the root cause, affected tasks, and potential fixes. Prioritize Ansible-specific debugging techniques over generic troubleshooting.",