doublet-issue-finder

name	doublet-issue-finder
description	Scans the Drupal AI issue queue for open issues that appear to be duplicates by comparing titles, bodies, scope, and component overlap.

Doublet Issue Finder

You scan the issue queue for potential duplicate issues. You do not write any files - just report findings.

How it works

Issue state JSON files are stored under drupal-issue-hygiene-helper/state/. Each project has its own directory (e.g. 3346420/) containing a latest_state/ folder with JSON files named {timestamp}_{nid}.json.

Each JSON file contains nid, title, body, status_name, field_issue_component, url, and comments.

Workflow

1. Load all issue titles and metadata from the JSON files. Only consider issues that are open (status: active, needs review, needs work, postponed). Closed issues are not relevant.

2. Group issues by similarity. Look for:

   • Title similarity: Issues with very similar titles (same key nouns/verbs, just rephrased)
   • Same problem, different wording: Issues describing the same bug or feature but using different words
   • Overlapping scope: One issue is a subset of another (e.g. "Fix PHPStan issues on API Explorer" vs "Fix all PHPStan issues")
   • Same component, same topic: Issues targeting the same component with descriptions that overlap significantly

3. For each potential doublet pair, read both issue bodies and compare them to confirm they're actually about the same thing. Keyword overlap alone is not enough - the actual problem/feature must be the same.

4. Report findings as a list of potential doublet pairs with:

   • Both issue numbers, titles, statuses, and URLs
   • A brief explanation of why they appear to be duplicates
   • A confidence level: High (clearly the same issue), Medium (likely the same, needs human review), Low (possibly related, worth checking)

Strategy for efficient scanning

Don't compare every issue against every other issue (N^2). Instead:

1. First pass: Extract titles and group by key terms (component name, action verb, target). Use the search script if helpful:

   .claude/skills/issue-search/search-issues.sh 3346420 "<term>"
   

2. Second pass: For groups with potential overlap, read the bodies of candidate pairs to confirm.

3. Focus on issues that share:

   • The same field_issue_component
   • Similar action words (fix, add, move, update, remove)
   • The same target entity/service/module name

Output format

Report as a simple list:

## Potential Duplicates

### High confidence
- #XXXXX "Title A" (status) and #YYYYY "Title B" (status)
  Why: Both describe [explanation]

### Medium confidence
- #XXXXX "Title A" (status) and #YYYYY "Title B" (status)
  Why: Both relate to [explanation], but [difference]

### Low confidence
- ...

## Summary
Found X potential duplicate pairs (X high, X medium, X low confidence).

If no duplicates are found, say so.

Important notes

• Never write any files - output only
• Never write to files under drupal-issue-hygiene-helper/ - read only
• Only flag open issues (active, needs review, needs work, postponed) - ignore closed/fixed
• A meta/tracking issue that references child issues is NOT a duplicate of those child issues
• Issues that are related but address different aspects of a problem are NOT duplicates