feat(ci): add duplicate issue detection and auto-close bot

[jquinter]

Summary

• Adds a Python script (.github/scripts/close_duplicate_issues.py) that detects duplicate issues using difflib.SequenceMatcher on normalized titles and closes them via the gh CLI
• Adds a workflow_dispatch workflow (scan_duplicate_issues.yml) for one-time batch scans of all open issues
• Extends the existing check_duplicate_issues.yml workflow to auto-close newly opened issues that are high-confidence duplicates

Two-tier duplicate detection

Threshold	Action
0.60	Informational comment + potential-duplicate label (existing wow-actions step)
0.85	Auto-close with comment, duplicate label, and not_planned state reason

Dry-run results against 830 open issues

• 0.70 threshold: 13 matches, 6 false positives (46%) — short titles like "support timezone setting" vs "support salt rotating" matched incorrectly
• 0.85 threshold: 7 matches, 0 false positives — all legitimate (empty junk titles at 100% or identical bug reports at 87%)

Test plan

• Ran dry-run scan locally: python3 .github/scripts/close_duplicate_issues.py --repo BerriAI/litellm --scan --threshold 0.85
• Trigger one-time scan workflow via gh workflow run scan_duplicate_issues.yml with close: false to verify in CI
• Re-run with close: true to close the 7 confirmed duplicates
• Test ongoing detection by creating a test issue with a known duplicate title

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Feb 28, 2026 3:19am

[greptile-apps]

Greptile Summary

Adds a two-tier duplicate issue detection system: the existing wow-actions/potential-duplicates step labels issues at 0.60 similarity, while a new Python script auto-closes issues at 0.85 similarity via the gh CLI. A new workflow_dispatch workflow enables one-time batch scans of all open issues.

• The core duplicate detection logic using difflib.SequenceMatcher at 0.85 threshold is well-tested (0 false positives against 830 issues) and includes dry-run safety by default
• check_duplicate_issues.yml is missing an explicit actions/setup-python step — the script uses Python 3.10+ syntax (str | None) and relies on the system Python, unlike the scan workflow which pins Python 3.11
• fetch_open_issues contains dead code (lines 44-47) that builds API parameters which are immediately overwritten
• For every newly opened issue, the workflow fetches all open issues (800+) to check for duplicates — this works but is worth noting for scalability

Confidence Score: 4/5

• This PR is safe to merge with minor improvements recommended — CI-only changes with no impact on application code.
• The PR adds CI tooling only (no changes to application code). The core detection logic is sound with a well-tested 0.85 threshold. The main concern is the missing actions/setup-python step in check_duplicate_issues.yml, which could cause failures if the runner's system Python drops below 3.10. The dead code and injection hardening suggestions are minor quality improvements.
• check_duplicate_issues.yml needs attention for the missing Python setup step. .github/scripts/close_duplicate_issues.py has dead code that should be cleaned up.

Important Files Changed

Filename	Overview
.github/scripts/close_duplicate_issues.py	New script for duplicate issue detection using difflib.SequenceMatcher. Contains dead code in fetch_open_issues (lines 44-47 overwritten by 48-49). Core logic is sound — threshold-based matching with dry-run safety. Minor issues only.
.github/workflows/check_duplicate_issues.yml	Extends existing workflow with auto-close steps gated on opened events. Missing actions/setup-python step — relies on system Python for PEP 604 syntax (`str
.github/workflows/scan_duplicate_issues.yml	New workflow_dispatch workflow for batch scanning. Properly sets up Python 3.11. Minor concern: ${{ inputs.threshold }} directly interpolated in shell (low risk since write-access gated).

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Issue Opened] --> B[wow-actions/potential-duplicates\nthreshold: 0.60]
    B --> C{Similarity >= 0.60?}
    C -->|Yes| D[Add 'potential-duplicate' label\n+ informational comment]
    C -->|No| E[No action]
    D --> F[Checkout close script]
    E --> F
    F --> G[Run close_duplicate_issues.py\nthreshold: 0.85]
    G --> H[Fetch ALL open issues via gh api]
    H --> I[Normalize titles & compare\nusing SequenceMatcher]
    I --> J{Similarity >= 0.85?}
    J -->|Yes| K[Add comment + 'duplicate' label\nClose with 'not_planned' reason]
    J -->|No| L[No further action]

    M[Manual Trigger\nworkflow_dispatch] --> N[Fetch ALL open issues]
    N --> O[Compare every issue\nagainst older issues]
    O --> P{Similarity >= threshold?}
    P -->|Yes & close=true| Q[Close as duplicate]
    P -->|Yes & close=false| R[Dry-run log only]
    P -->|No| S[Skip]

Loading

_{Last reviewed commit: db3d61f}

[greptile-apps]

_{3 files reviewed, 3 comments}

_{Edit Code Review Agent Settings | Greptile}

[shin-bot-litellm]

Review

1. Does this PR fix the issue it describes?
Yes. Adds a duplicate issue detection system with two tiers:

• 0.60 threshold → informational comment + label
• 0.85 threshold → auto-close (verified 0 false positives on 830 issues)

Uses difflib.SequenceMatcher on normalized titles. Well-documented dry-run results.

2. Has this issue already been solved elsewhere?
Partially — check_duplicate_issues.yml exists using wow-actions but only adds labels at 0.60. This extends it with auto-close at 0.85.

3. Are there other PRs addressing the same problem?
No duplicates found.

4. Are there other issues this potentially closes?
Will help reduce issue backlog noise — 7 confirmed duplicates identified in dry run.

✅ LGTM — conservative threshold (0.85) ensures no false positives. Good CI addition.