claude-issue-triage.yml

name: Issue Triage

on:
  issues:
    types: [opened]

jobs:
  triage:
    runs-on: ubuntu-latest
    timeout-minutes: 10
    permissions:
      contents: read
      issues: write
      id-token: write

    steps:
      - name: Checkout repository
        uses: actions/checkout@v4
        with:
          fetch-depth: 1

      - name: Triage and check for duplicates
        uses: anthropics/claude-code-action@v1
        with:
          anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
          github_token: ${{ secrets.GITHUB_TOKEN }}
          allowed_non_write_users: "*"
          prompt: |
            Analyze this new issue and perform triage tasks.

            Issue: #${{ github.event.issue.number }}
            Repository: ${{ github.repository }}

            ## Your Tasks:

            ### 1. Get issue details
            Use mcp__github__get_issue to get the full details of issue #${{ github.event.issue.number }}

            ### 2. Check for duplicates
            Search for similar existing issues using mcp__github__search_issues with relevant keywords from the issue title and body.

            Criteria for duplicates:
            - Same bug or error being reported
            - Same feature request (even if worded differently)
            - Same question being asked
            - Issues describing the same root problem

            If you find a duplicate:
            - Add a comment using EXACTLY this format (required for auto-close to work):
              "Found a possible duplicate of #<issue_number>: <brief explanation of why it's a duplicate>"
            - Do NOT apply the "duplicate" label yet (the auto-close script will add it after 12 hours if no objections)
            - Suggest the user react with a thumbs-down if they disagree

            ### 3. Check for Low-Quality / AI Spam
            Analyze the issue quality. We are receiving many low-effort, AI-generated spam issues.
            Flag the issue as INVALID if it matches these criteria:
            - **Vague/Generic**: Title is "Fix bug" or "Error" without specific context.
            - **Hallucinated**: Refers to files or features that do not exist in this repo.
            - **Template Filler**: Body contains "Insert description here" or unrelated gibberish.
            - **Low Effort**: No reproduction steps, no logs, only 1-2 sentences.

            If identified as spam/low-quality:
            - Add the "invalid" label.
            - Add a comment:
              "This issue has been automatically flagged as low-quality or potentially AI-generated spam. It lacks specific details (logs, reproduction steps, file references) required for us to help. Please open a new issue following the template exactly if this is a legitimate request."
            - Do NOT proceed to other steps.

            ### 4. Check for invalid issues (General)
            If the issue is not spam but still lacks information:
            - Add the "invalid" label
            - Comment asking for clarification

            ### 5. Categorize with labels (if NOT a duplicate or spam)
            Apply appropriate labels based on the issue content. Use ONLY these labels:
            - bug: Something isn't working
            - enhancement: New feature or request
            - question: Further information is requested
            - documentation: Improvements or additions to documentation
            - good first issue: Good for newcomers (if issue is well-defined and small scope)
            - help wanted: Extra attention is needed (if issue needs community input)
            - backlog: Tracked for the future, but not currently planned or prioritized

            ### 6. Estimate size (if NOT a duplicate, spam, or invalid)
            Apply exactly ONE size label to help contributors match their capacity to the task:
            - "size: small": Docs, typos, single-file fixes, config changes
            - "size: medium": Bug fixes with tests, adding a single tool, changes within one package
            - "size: large": Cross-package changes (core + tools), new modules, complex logic, architectural refactors

            You may apply multiple labels if appropriate (e.g., "bug", "size: small", and "good first issue").

            ## Tools Available:
            - mcp__github__get_issue: Get issue details
            - mcp__github__search_issues: Search for similar issues
            - mcp__github__list_issues: List recent issues if needed
            - mcp__github__add_issue_comment: Add a comment
            - mcp__github__update_issue: Add labels
            - mcp__github__get_issue_comments: Get existing comments

            Be thorough but efficient. Focus on accurate categorization and finding true duplicates.

          claude_args: |
            --model claude-haiku-4-5-20251001
            --allowedTools "mcp__github__get_issue,mcp__github__search_issues,mcp__github__list_issues,mcp__github__add_issue_comment,mcp__github__update_issue,mcp__github__get_issue_comments"