Issue Labeler

Generate label recommendations for unlabeled GitHub issues, review them, then submit in batch.

When to Use

• Repository has many unlabeled or poorly labeled issues
• You want to triage issues without manually reading each one
• Need consistent labeling based on issue content

Workflow

Step 1: Generate Recommendations

# Fetch unlabeled issues and analyze them
./scripts/analyze.sh owner/repo

This creates recommendations.json with suggested labels for each issue.

Step 2: Review Recommendations

# Launch the review UI
./scripts/serve.sh

The UI shows:

• Issue title and preview
• Current labels (if any)
• Recommended labels with confidence
• Checkbox to approve/reject each recommendation

Step 3: Submit Approved Labels

After reviewing in the UI, export approved recommendations and run:

# Apply approved labels
./scripts/apply.sh recommendations-approved.json

Or use the "Submit All" button in the UI to apply via gh CLI.

Label Categories

The analyzer suggests labels from these categories:

How Analysis Works

For each unlabeled issue, the LLM analyzes:

1. Title keywords - error, feature, how to, crash, etc.
2. Body content - stack traces, repro steps, feature requests
3. Existing patterns - what labels similar issues have
4. Repository context - available labels in the repo

Files

issue-labeler/
├── SKILL.md              # This file
├── review.html           # Review UI for recommendations
├── scripts/
│   ├── analyze.sh        # Fetch and generate recommendations
│   ├── apply.sh          # Apply approved labels
│   └── serve.sh          # Launch review UI
├── recommendations.json  # Generated recommendations (git-ignored)
└── approved.json         # Approved recommendations (git-ignored)

Example Recommendation

{
  "number": 123,
  "title": "App crashes when clicking submit",
  "current_labels": [],
  "recommended_labels": ["bug", "priority-2", "needs-info"],
  "confidence": 0.85,
  "reasoning": "Title indicates crash (bug). No repro steps provided (needs-info). User-facing issue (priority-2).",
  "approved": null
}

Safety

• No labels applied without explicit approval
• Review UI shows reasoning for each recommendation
• Dry-run mode available: ./scripts/apply.sh --dry-run
• All actions logged for audit

Notes

• Requires gh CLI authenticated with write access to issues
• Run gh auth status to verify permissions
• For large repos, analyze in batches using --limit flag