Gha Issue Triage — AI skill for Claude Code

gha-issue-triage

  [](https://github.com/qte77/gha-issue-triage/actions/workflows/test.yml)  [](https://github.com/qte77/gha-issue-triage/actions/workflows/dependabot/dependabot-updates) [](https://github.com/qte77/gha-issue-triage/actions/workflows/ruff.yml)

AI-powered issue triage GitHub Action. Detects duplicates, scores relevance, analyzes feasibility, auto-labels, and posts a sticky summary comment with the analysis (edited in place on re-runs).

What it does

1. Duplicate Detection — Fuzzy matches new issues against existing ones using difflib.SequenceMatcher
2. Relevance Scoring — LLM-based scoring against repo scope (README.md, CLAUDE.md)
3. Feasibility Analysis — Two orthogonal judgements per issue:
   • feasibility (yes / no) — can this be built at all? (no means out-of-physics / out-of-scope of software, e.g. "build a faster-than-light drive".)
   • complexity (low / medium / high) — if feasible, how hard? Drives good first issue when low.
4. Auto-Labeling — Applies labels (aligned with GitHub's default label set): bug, documentation, duplicate, enhancement, feature, good first issue, invalid, needs discussion
5. Sticky Summary Comment — Posts a single bot comment with the analysis (relevance, feasibility, duplicate match). Re-runs edit the same comment instead of stacking new ones. On auth/API failures (missing models: read, expired PAT, fork-PR read-only token, rate limit, etc.), the