CodelldBot

.github/workflows/analyze_issue.py

@@ -0,0 +1,150 @@
+import os
+import json
+import io
+import time
+from openai import OpenAI
+from openai.types.beta.assistant_stream_event import ThreadRunRequiresAction, ThreadMessageCompleted
+from octokit import Octokit
+
+class IssueAnalyzer:
+    def __init__(self):
+        self.octokit = Octokit()
+        self.openai = OpenAI()
+        self.repo_full_name = os.getenv('GITHUB_REPOSITORY')
+
+    def handle_event(self):
+
+        with open(os.getenv('GITHUB_EVENT_PATH'), 'rb') as f:
+            event = json.load(f)
+
+        match os.getenv('GITHUB_EVENT_NAME'):
+            case 'issues':
+                issue = event['issue']
+            case 'workflow_dispatch':
+                issue_number = int(event['inputs']['issue'])
+                owner, repo = self.repo_full_name.split('/')
+                response = self.octokit.issues.get(owner=owner, repo=repo, issue_number=issue_number)
+                issue = response.json
+
+        assistant = self.openai.beta.assistants.retrieve(os.getenv('ASSISTANT_ID'))
+
+        issue_file = self.openai.files.create(
+            file=('BUG_REPORT.md', self.make_issue_content(issue, show_labels=False)),
+            purpose='assistants'
+        )
+
+        thread = self.openai.beta.threads.create(
+            metadata={
+                'issue': f'{issue["number"]}: {issue["title"]}',
+                'run_id': os.getenv('GITHUB_RUN_ID'),
+                'model': f'{assistant.model} t={assistant.temperature} top_p={assistant.top_p}',
+            },
+            messages=[{
+                    'role': 'user',
+                    'content': 'We have a new issue report (attached as BUG_REPORT.md)',
+                    'attachments': [{
+                        'file_id': issue_file.id,
+                        'tools': [{'type': 'file_search'}]
+                    }]
+                }
+            ]
+        )
+        print('Thread:', thread.id)
+
+        thread_vstore_id = thread.tool_resources.file_search.vector_store_ids[0]
+        self.wait_vector_store(thread_vstore_id)
+
+        stream = self.openai.beta.threads.runs.create(
+            assistant_id=assistant.id,
+            thread_id=thread.id,
+            stream=True
+        )
+
+        streams = [stream]
+        while streams:
+            stream = streams.pop(0)
+            for event in stream:
+                match event:
+                    case ThreadMessageCompleted():
+                        for c in event.data.content:
+                            print('<< Message >>', c.text.value)
+                    case ThreadRunRequiresAction():
+                        tool_outputs = []
+                        for tool in event.data.required_action.submit_tool_outputs.tool_calls:
+                            args = json.loads(tool.function.arguments)
+                            print(f'<< Tool call >>', tool.function.name, args)
+                            match tool.function.name:
+                                case 'search_github':
+                                    query = f'repo:{self.repo_full_name} {args["query"]}'
+                                    output = self.search_github(query, thread_vstore_id, exclude=[issue['number']])
+                                    tool_outputs.append({'tool_call_id': tool.id, 'output': output})
+                                case 'add_issue_labels':
+                                    tool_outputs.append({'tool_call_id': tool.id, 'output': 'Ok'})
+                                case 'set_issue_title':
+                                    tool_outputs.append({'tool_call_id': tool.id, 'output': 'Ok'})
+                                case 'add_issue_comment':
+                                    tool_outputs.append({'tool_call_id': tool.id, 'output': 'Ok'})
+
+                        new_stream = self.openai.beta.threads.runs.submit_tool_outputs(
+                            thread_id=thread.id,
+                            run_id=event.data.id,
+                            tool_outputs=tool_outputs,
+                            stream=True)
+                        streams.append(new_stream)
+
+    def search_github(self, query: str, vstore_id: str, exclude:list=[], max_results=5) -> str:
+        response = self.octokit.search.issues(q=query)
+        if response.json.get('status'):
+            return f'Search failed: {response.json["message"]}'
+
+        result_lines = []
+        for issue in response.json['items']:
+            issue_number = issue['number']
+            if issue_number in exclude:
+                continue
+            issue_file = self.openai.files.create(
+                file=(f'ISSUE_{issue_number}.md', self.make_issue_content(issue, fetch_comments=True)),
+                purpose='assistants'
+            )
+            self.openai.beta.vector_stores.files.create(
+                vector_store_id=vstore_id,
+                file_id=issue_file.id,
+            )
+            result_lines.append(f'Issue number: {issue_number}, file name: {issue_file.filename}')
+            if len(result_lines) >= max_results:
+                break
+
+        self.wait_vector_store(vstore_id)
+
+        result_lines.insert(0, f'Found {len(result_lines)} issues and attached as files to this thread:')
+        return '\n'.join(result_lines)
+
+    def make_issue_content(self, issue, fetch_comments=False, show_labels=True) -> bytes:
+        f = io.StringIO()
+        f.write(f'### Title: {issue["title"]}\n')
+        f.write(f'### Author: {issue["user"]["login"]}\n')
+        f.write(f'### State: {issue["state"]}\n')
+        if show_labels:
+            f.write(f'### Labels: {",".join(label["name"] for label in issue["labels"])}\n')
+        f.write(f'\n{issue["body"]}\n')
+
+        if fetch_comments:
+            owner, repo = self.repo_full_name.split('/')
+            comments = self.octokit.issues.list_issue_comments(
+                owner=owner, repo=repo, issue_number=issue['number'])
+            for comment in comments.json:
+                f.write(f'### Comment by {comment["user"]["login"]}\n')
+                f.write(f'\n{comment["body"]}\n')
+
+        return f.getvalue().encode('utf-8')
+
+    def wait_vector_store(self, vstore_id):
+        vstore = self.openai.beta.vector_stores.retrieve(vstore_id)
+        while vstore.status == 'in_progress':
+            print('Waiting for vector store update.')
+            time.sleep(1)
+            vstore = self.openai.beta.vector_stores.retrieve(vstore_id)
+
+
+if __name__ == '__main__':
+    IssueAnalyzer().handle_event()

.github/workflows/analyze_issue.yml

@@ -0,0 +1,35 @@
+name: analyze_issue
+
+on:
+  issues:
+    types: [opened]
+  workflow_dispatch:
+    inputs:
+      issue: null
+
+jobs:
+  analyze_issue:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Check out repo
+        uses: actions/checkout@v3
+
+      - name: Set up Python
+        uses: actions/setup-python@v4
+        with:
+          python-version: "3.11"
+          cache: 'pip'
+          cache-dependency-path: .github/workflows/analyze_issue.yml
+
+      - name: Install dependencies
+        run: pip install openai octokitpy
+
+      - name: Analysis
+        env:
+          GITHUB_TOKEN: ${{ github.token }}
+          GITHUB_REPOSITORY: ${{ github.repository }}
+          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
+          ASSISTANT_ID: ${{ vars.ASSISTANT_ID }}
+          PYTHONUNBUFFERED: 1
+        run: |
+          python .github/workflows/analyze_issue.py