Move prompts to a YAML file, add different style

pyproject.toml

@@ -13,6 +13,7 @@ dependencies = [
     "PyGithub==2.4.0",
     "ollama==0.3.3",
     "colorama==0.4.6",
+    "pyyaml==6.0.1",
 ]
 
 [project.urls]

src/agents.py

@@ -1,18 +1,27 @@
 import ollama
+import yaml
 
 from model import post_request_generate
-from prompts import CODE_REVIEW_PROMPT, FEEDBACK_IMPROVEMENT_PROMPT, EXAMPLES
 
 class Agent:
-    def __init__(self, model_name="llama3.1", role="code_reviewer", use_post_request=False):
+    def __init__(self, model_name="llama3.1", role="code_reviewer", use_post_request=False, prompt_style="default"):
         self.model_name = model_name
         self.role = role
         self.use_post_request = use_post_request
+        self.prompt_style = prompt_style
+        self.prompts = self._load_prompts()
+        print(f"Inited agent with model: {model_name}, role: {role}, use_post_request: {use_post_request}, prompt_style: {prompt_style}")
+
+    def _load_prompts(self):
+        with open('src/prompts.yaml', 'r') as file:
+            return yaml.safe_load(file)['prompts']
 
     def analyze(self, data, model_name=None):
         prompt = self._prepare_prompt(data)
         model = model_name or self.model_name
 
+        # print(f"Prompt:\n{prompt}\nENDOFPROMPT")
+
         if self.use_post_request:
             response = post_request_generate(model, prompt)
         else:
@@ -25,19 +34,19 @@ def _prepare_prompt(self, data):
             return self._prepare_code_review_prompt(data)
         elif self.role == "feedback_improver":
             prompt = self._prepare_feedback_improvement_prompt(data)
-            print(f"IMPROVERPROMPT:\n{prompt}\nENDOFIMPROVERPROMPT")
+            # print(f"IMPROVERPROMPT:\n{prompt}\nENDOFIMPROVERPROMPT")
             return prompt
         else:
             raise ValueError(f"Unknown role: {self.role}")
 
     def _prepare_code_review_prompt(self, pr_data):
         context = f"Additional Context:\n{pr_data['context']}" if pr_data.get('context') else ""
-        return CODE_REVIEW_PROMPT.format(
+        return self.prompts['instructions'][self.prompt_style]['review'].format(
             title=pr_data['title'],
             description=pr_data['description'],
             changes=pr_data['changes'],
             context=context,
-            examples=EXAMPLES
+            examples=self.prompts['examples']
         )
 
     def _prepare_feedback_improvement_prompt(self, review_data):
@@ -50,11 +59,11 @@ def _prepare_feedback_improvement_prompt(self, review_data):
 {review}
 ---
 """
-        return FEEDBACK_IMPROVEMENT_PROMPT.format(
+        return self.prompts['instructions'][self.prompt_style]['feedback'].format(
             title=review_data['pr_data']['title'],
             description=review_data['pr_data']['description'],
             changes=review_data['pr_data']['changes'],
             reviews=reviews,
             context=context,
-            examples=EXAMPLES
-        )
+            examples=self.prompts['examples']
+        )

src/pearbot.py

@@ -16,6 +16,7 @@ def main():
     parser.add_argument("--server", action="store_true", help="Run as a server")
     parser.add_argument("--diff", type=str, nargs='?', const='-', help="Path to the diff file or '-' for stdin")
     parser.add_argument("--model", type=str, default="llama3.1", help="Model for the final review step")
+    parser.add_argument("--prompt-style", type=str, default="default", help="Prompt style (from prompts.yaml)")
     parser.add_argument("--initial-review-models", type=str, default="llama3.1,llama3.1,llama3.1", help="Comma-separated list of model names for the initial review (default: llama3.1,llama3.1,llama3.1)")
 
     args = parser.parse_args()
@@ -25,8 +26,8 @@ def main():
 
     print(f"Available models: {', '.join(get_available_models()) or 'NONE'}")
 
-    code_review_agent = Agent(role="code_reviewer", use_post_request=True)
-    feedback_improver_agent = Agent(role="feedback_improver", use_post_request=True)
+    code_review_agent = Agent(role="code_reviewer", use_post_request=True, prompt_style=args.prompt_style)
+    feedback_improver_agent = Agent(role="feedback_improver", use_post_request=True, prompt_style=args.prompt_style)
 
     if args.server:
         print("Running as a server...")

src/prompts.yaml

@@ -0,0 +1,129 @@
+prompts:
+  examples: |
+    Example #1:
+    Change:
+    ```diff
+    + fi
+
+    +if [[ $2 =~ ^[0-9]+$ ]]; then
+    +    msgSize=$1
+    ```
+    Suggestion:
+    The assignment of msgSize is incorrect; it should be assigned the value of $2 instead of $1:
+    ```diff
+    -    msgSize=$1
+    +    msgSize=$2
+    ```
+
+    Example #2:
+    Change:
+    ```diff
+    + xterm -geometry 90x40+550+40 -hold -e @EX_BIN_DIR@/$PROCESSOR1 &
+    +
+    + PROCESSOR2="fairmq-ex-region-processor"
+    + PROCESSOR2+=" --id processor1"
+    ```
+    Suggestion:
+    There seems to be a copy-paste error. The `PROCESSOR2` should have a unique identifier, but it is currently set to `processor1`, which is the same as `PROCESSOR1`. This should be corrected to ensure that each processor has a unique ID.
+    ```diff
+    -    msgSize=$1
+    +    msgSize=$2
+    ```
+
+  instructions:
+    default:
+      review: |
+        You are an experienced software engineer tasked with reviewing the following Pull Request.
+
+        ---
+        Pull Request title: {title}
+        Pull Request description:
+        {description}
+        ---
+
+        Analyze the following file changes:
+
+        ---
+        {changes}
+        ---
+
+        {context}
+
+        Please provide specific comments only for changed lines (additions or deletions), where you see potential issues, improvements or suggestions for the developer.
+        Take into account the existing comments and avoid repeating already mentioned points.
+        Separate your comments by file and provide a clear and concise explanation for each comment.
+        Don't repeat or mention the instructions given to you, just perform them.
+        Reference the relevant diff lines and keep your suggestions short and concise as in these examples:
+
+        ---
+        {examples}
+        ---
+
+        DO NOT include or comment on the examples in your reply. Your response:
+            Let's work this out in a step by step way to be sure we provide only useful suggestions:
+
+      feedback: |
+        As an expert software engineer, you are tasked with providing feedback for a Pull Request.
+        Consider the original Pull Request details and a number of preliminary reviews.
+
+        Original Pull Request:
+        ---
+        Title: {title}
+        Description:
+        {description}
+        ---
+
+        With the following file changes:
+
+        ---
+        {changes}
+        ---
+
+        {context}
+
+        Preliminary Reviews:
+        ---
+        {reviews}
+        ---
+
+        Based on these reviews, but also adding your expertise on top, provide a review that contains feedback for the important aspects of the code changes.
+        Include only points that could potentially fix errors or lead to improvements.
+        DO NOT comment on the quality of the preliminary reviews themselves and DO NOT quote the preliminary reviews, but address the points directly.
+
+        Reference the relevant diff lines and keep your suggestions short and concise as in these examples:
+
+        ---
+        {examples}
+        ---
+
+        DO NOT include or comment on the examples in your reply. Only your feedback to the Pull Request code changes:
+
+    simple:
+      review: |
+        Review the following Pull Request:
+
+        Title: {title}
+        Description: {description}
+
+        Changes:
+        {changes}
+
+        {context}
+
+        Provide concise feedback on potential issues or improvements. Focus on changed lines only.
+
+      feedback: |
+        Improve the feedback for this Pull Request:
+
+        Title: {title}
+        Description: {description}
+
+        Changes:
+        {changes}
+
+        {context}
+
+        Previous reviews:
+        {reviews}
+
+        Provide a concise, improved review focusing on the most important aspects of the code changes. Consider preliminary reviews as reference, but do not quote or mention them in any way.