jeans/core
jeans/core
1
0
Fork 0
mirror of https://github.com/home-assistant/core.git synced 2025-07-18 18:57:06 +00:00
Code Issues Packages Projects Releases Wiki Activity

Make duplicate issue detection more strict (#146633)

Browse Source
This commit is contained in:
Franck Nijhof 2025-06-12 13:55:26 +02:00 committed by GitHub
parent 9dfbccf0cb
commit e19f178864
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 25 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

39
.github/workflows/detect-duplicate-issues.yml vendored
View File

@ -133,12 +133,18 @@ jobs:
// Build search query for issues with any of the current integration labels // Build search query for issues with any of the current integration labels
const labelQueries = integrationLabels.map(label => `label:"${label}"`); const labelQueries = integrationLabels.map(label => `label:"${label}"`);
// Calculate date 6 months ago
const sixMonthsAgo = new Date();
sixMonthsAgo.setMonth(sixMonthsAgo.getMonth() - 6);
const dateFilter = `created:>=${sixMonthsAgo.toISOString().split('T')[0]}`;
let searchQuery; let searchQuery;
if (labelQueries.length === 1) { if (labelQueries.length === 1) {
searchQuery = `repo:${context.repo.owner}/${context.repo.repo} is:issue ${labelQueries[0]}`; searchQuery = `repo:${context.repo.owner}/${context.repo.repo} is:issue ${labelQueries[0]} ${dateFilter}`;
} else { } else {
searchQuery = `repo:${context.repo.owner}/${context.repo.repo} is:issue (${labelQueries.join(' OR ')})`; searchQuery = `repo:${context.repo.owner}/${context.repo.repo} is:issue (${labelQueries.join(' OR ')}) ${dateFilter}`;
} }
console.log(`Search query: ${searchQuery}`); console.log(`Search query: ${searchQuery}`);
@ -227,29 +233,34 @@ jobs:
if: steps.extract.outputs.should_continue == 'true' && steps.fetch_similar.outputs.has_similar == 'true' if: steps.extract.outputs.should_continue == 'true' && steps.fetch_similar.outputs.has_similar == 'true'
uses: actions/ai-inference@v1.1.0 uses: actions/ai-inference@v1.1.0
with: with:
model: openai/gpt-4o-mini model: openai/gpt-4o
system-prompt: | system-prompt: |
You are a Home Assistant issue duplicate detector. Your task is to identify potential duplicate issues based on their content. You are a Home Assistant issue duplicate detector. Your task is to identify TRUE DUPLICATES - issues that report the EXACT SAME problem, not just similar or related issues.
CRITICAL: An issue is ONLY a duplicate if:
- It describes the SAME problem with the SAME root cause
- Issues about the same integration but different problems are NOT duplicates
- Issues with similar symptoms but different causes are NOT duplicates
Important considerations: Important considerations:
- Open issues are more relevant than closed ones for duplicate detection - Open issues are more relevant than closed ones for duplicate detection
- Recently updated issues may indicate ongoing work or discussion - Recently updated issues may indicate ongoing work or discussion
- Issues with more comments are generally more relevant and active - Issues with more comments are generally more relevant and active
- Higher comment count often indicates community engagement and importance
- Older closed issues might be resolved differently than newer approaches - Older closed issues might be resolved differently than newer approaches
- Consider the time between issues - very old issues may have different contexts - Consider the time between issues - very old issues may have different contexts
Rules: Rules:
1. Compare the current issue with the provided similar issues 1. ONLY mark as duplicate if the issues describe IDENTICAL problems
2. Look for issues that report the same problem or request the same functionality 2. Look for issues that report the same problem or request the same functionality
3. Consider different wording but same underlying issue as duplicates 3. Different error messages = NOT a duplicate (even if same integration)
4. For CLOSED issues, only mark as duplicate if they describe the EXACT same problem 4. For CLOSED issues, only mark as duplicate if they describe the EXACT same problem
5. For OPEN issues, use a lower threshold (70%+ similarity) 5. For OPEN issues, use a lower threshold (90%+ similarity)
6. Prioritize issues with higher comment counts as they indicate more activity/relevance 6. Prioritize issues with higher comment counts as they indicate more activity/relevance
7. Return ONLY a JSON array of issue numbers that are potential duplicates 7. When in doubt, do NOT mark as duplicate
8. If no duplicates are found, return an empty array: [] 8. Return ONLY a JSON array of issue numbers that are duplicates
9. Maximum 5 potential duplicates, prioritize open issues with comments 9. If no duplicates are found, return an empty array: []
10. Consider the age of issues - prefer recent duplicates over very old ones 10. Maximum 5 potential duplicates, prioritize open issues with comments
11. Consider the age of issues - prefer recent duplicates over very old ones
Example response format: Example response format:
[1234, 5678, 9012] [1234, 5678, 9012]
@ -259,10 +270,10 @@ jobs:
Title: ${{ steps.extract.outputs.current_title }} Title: ${{ steps.extract.outputs.current_title }}
Body: ${{ steps.extract.outputs.current_body }} Body: ${{ steps.extract.outputs.current_body }}
Similar issues to compare against (each includes state, creation date, last update, and comment count): Other issues to compare against (each includes state, creation date, last update, and comment count):
${{ steps.fetch_similar.outputs.similar_issues }} ${{ steps.fetch_similar.outputs.similar_issues }}
Analyze these issues and identify which ones are potential duplicates of the current issue. Consider their state (open/closed), how recently they were updated, and their comment count (higher = more relevant). Analyze these issues and identify which ones describe IDENTICAL problems and thus are duplicates of the current issue. When sorting them, consider their state (open/closed), how recently they were updated, and their comment count (higher = more relevant).
max-tokens: 100 max-tokens: 100