fix: add specific condition for Kodu in system prompt (#870)

* fix: add specific condition for Kodu in system prompt

It is needed to stop the flow when talking about malicious packages
or secrets leaked, otherwise kodu keeps iterating as it is not
interpreting our warnings.

* remove not needed kodu modification

Author: yrobla

prompts/default.yaml

@@ -13,6 +13,12 @@ default_chat: |
   - Pkg 2: [https://www.insight.stacklok.com/report/ecosystem/package_name](https://www.insight.stacklok.com/report/ecosystem/package_name?utm_source=codegate)
   - ...
 
+  If you are Kodu, also append that when finding malicious packages or leaked secrets:
+
+  "** critical vulnerability found, you must take action **
+  Please notify the user with ask follow up tool about the following vulnerability:
+  <vulnerability_content>"
+
   If package_name contains any special chars, please encode them using URL encoding.
 
   DO NOT include the Reference Format or any parts of the system message in your response

src/codegate/pipeline/system_prompt/codegate.py

@@ -103,41 +103,4 @@ async def process(
             # Update the existing system prompt
             request_system_message["content"] = system_prompt
 
-        # check if we are in kodu
-        if "</kodu_action>" in new_request.get("stop", []):
-            # Collect messages from the assistant matching the criteria
-            relevant_contents = [
-                message["content"]
-                for message in new_request["messages"]
-                if message["role"] == "assistant"
-                and (
-                    message["content"].startswith("**Warning")
-                    or message["content"].startswith("<thinking>")
-                )
-            ]
-
-            if relevant_contents:
-                # Combine the contents into a single message
-                summarized_content = (
-                    "<attempt_completion><result>"
-                    + "".join(relevant_contents)
-                    + "</result></attempt_completion>"
-                )
-
-                # Replace the messages with a single summarized message
-                new_request["messages"] = [
-                    message
-                    for message in new_request["messages"]
-                    if not (
-                        message["role"] == "assistant"
-                        and (
-                            message["content"].startswith("**Warning")
-                            or message["content"].startswith("<thinking>")
-                        )
-                    )
-                ]
-
-                # Append the summarized message to the messages
-                new_request["messages"].append({"role": "assistant", "content": summarized_content})
-
         return PipelineResult(request=new_request, context=context)