GitHub Issue Dedupe

Overview

Automatically detect duplicate GitHub issues using semantic similarity, link them together, close duplicates with a reference comment, and maintain a canonical issue with consolidated context.

When to Use

• •When your issue tracker has grown large with many potential duplicates
• •Before a sprint planning session to clean up the backlog
• •After a major release when bug reports flood in
• •When you want to prevent duplicate effort on the same problem
• •When enforcing issue hygiene across a public open-source project

Instructions

1. •Fetch all open issues: gh issue list --state open --limit 500 --json number,title,body,labels.
2. •Build a similarity matrix:
   • •Tokenize titles and bodies, remove stop words.
   • •Compute cosine similarity between all issue pairs.
   • •Flag pairs exceeding DUPLICATE_THRESHOLD as potential duplicates.
3. •For each duplicate cluster:
   • •Identify the canonical issue (oldest or highest vote count).
   • •Review pairs above threshold to confirm semantic similarity.
   • •Post comment on duplicate: "Duplicate of #<canonical> - closing in favor of original."
   • •Close duplicate with label duplicate.
   • •Update canonical issue body with any unique details from duplicates.
4. •Handle false positives: skip pairs where titles differ by key identifiers (version numbers, OS, component).
5. •Return deduplication report: total issues scanned, duplicates found, clusters merged.

Environment

DUPLICATE_THRESHOLD=0.82
BATCH_SIZE=100
DRY_RUN=false
SKIP_LABELS=wontfix,known-issue
MERGE_COMMENTS=true

Examples

Input:

repository: org/my-app
state: open
limit: 200

Output:

Deduplication Report
Issues scanned: 187
Duplicate clusters found: 14
Issues closed: 23
Sample merges:
- #445, #467, #489 -> canonical: #445 ("Login button not working")
- #312, #398 -> canonical: #312 ("PDF export fails on Safari")
False positives skipped: 3