https://github.com/qte77/gha-issue-triage

AI-powered issue triage GitHub Action: duplicate detection, relevance scoring, feasibility analysis, auto-labeling, and a sticky summary comment per issue. Backends: GitHub Models, Anthropic, OpenAI-compatible.
https://github.com/qte77/gha-issue-triage

actions ai anthropic auto-labeling automation claude composite-action duplicate-detection github-action github-models issue-triage llm

Last synced: 13 days ago
JSON representation

Host: GitHub
URL: https://github.com/qte77/gha-issue-triage
Owner: qte77
License: apache-2.0
Created: 2026-04-06T22:16:07.000Z (2 months ago)
Default Branch: main
Last Pushed: 2026-05-18T21:50:11.000Z (28 days ago)
Last Synced: 2026-05-18T22:37:26.548Z (28 days ago)
Topics: actions, ai, anthropic, auto-labeling, automation, claude, composite-action, duplicate-detection, github-action, github-models, issue-triage, llm
Language: Python
Homepage:
Size: 458 KB
Stars: 2
Watchers: 0
Forks: 4
Open Issues: 14
Metadata Files:
- Readme: README.md
- Changelog: CHANGELOG.md
- License: LICENSE

Awesome Lists containing this project

README

          # gha-issue-triage

![Version](https://img.shields.io/badge/version-0.3.0-8A2BE2)

![License](https://img.shields.io/badge/license-Apache--2.0-blue)

[![Tests](https://github.com/qte77/gha-issue-triage/actions/workflows/test.yml/badge.svg)](https://github.com/qte77/gha-issue-triage/actions/workflows/test.yml)

![CodeFactor](https://www.codefactor.io/repository/github/qte77/gha-issue-triage/badge)

[![Dependabot Updates](https://github.com/qte77/gha-issue-triage/actions/workflows/dependabot/dependabot-updates/badge.svg)](https://github.com/qte77/gha-issue-triage/actions/workflows/dependabot/dependabot-updates)

[![Ruff](https://github.com/qte77/gha-issue-triage/actions/workflows/ruff.yml/badge.svg)](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 same sticky slot carries a `### Triage failure` comment with the concrete fix — see [`docs/integrations.md#troubleshooting`](docs/integrations.md#troubleshooting).

Screenshot — triaged issues in this repo

  

  

## Inputs

| Name | Required | Default | Description |

| --- | --- | --- | --- |

| `GH_TOKEN` | Yes | — | GitHub token for gh CLI |

| `AI_TOKEN` | No | `github.token` | GitHub Models API token. Default works as long as the caller workflow declares `permissions: models: read`. See [`docs/integrations.md`](docs/integrations.md) for PAT alternatives. |

| `MODEL` | No | `openai/gpt-4.1` | LLM model |

| `ANTHROPIC_API_KEY` | No | — | Anthropic API key (alternative backend) |

| `OPENAI_API_BASE` | No | — | Base URL of an OpenAI-compatible endpoint (Mistral/Ollama/vLLM); takes precedence when set |

| `MAX_DUPLICATES` | No | `10` | Max duplicate candidates |

| `SIMILARITY_THRESHOLD` | No | `0.6` | Fuzzy match threshold (0-1) |

## Usage

```yaml

name: Issue Triage

on:

  issues:

    types: [opened, edited]

jobs:

  triage:

    runs-on: ubuntu-latest

    permissions:

      issues: write

      contents: read

    steps:

      - uses: qte77/gha-issue-triage@4a07dd23bdd6bafc625bce6430f0aa5990fc327d  # v0.3.0

        with:

          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}

```

## Try it in this repo

This repo dogfoods the action via [`.github/workflows/self-triage.yml`](.github/workflows/self-triage.yml). Every new or edited issue is triaged automatically — no opt-in needed. Side effects: labels may be added, and one sticky summary comment is posted (edited in place on re-runs).

### Sample summary comment

```md

### AI triage summary

- **Duplicate of:** #30 (similarity 0.93)

- **Relevance:** 2/10 — `invalid` — The issue proposes an unrealistic feature unrelated to the repository's scope.

- **Feasibility:** `no` — Faster-than-light travel violates known physics.

```

When `feasibility` is `yes` the comment also shows a `Complexity:` line:

```md

- **Feasibility:** `yes`

- **Complexity:** `medium` — Requires extending the event-handler and adding tests. (~days)

```

The duplicate line is omitted when no duplicate is found.

## Choosing a model

`MODEL` defaults to `openai/gpt-4.1` (GitHub Models). Issue triage is a small/fast model workload — swap to a cheaper or faster model with a one-line caller change. No code change required.

| Use case | Suggested `MODEL` |

| --- | --- |

| Default — strongest general model on free tier | `openai/gpt-4.1` |

| Speed/cost balance | `openai/gpt-4o-mini` |

| Highest throughput | `microsoft/phi-4-mini-instruct` |

| Code-heavy repo (better feasibility scoring) | `deepseek/deepseek-v3-0324` |

| Open-weights preference | `meta/llama-4-scout-17b-16e-instruct` |

See [`docs/integrations.md`](docs/integrations.md) for the full catalog, rationale, and per-model notes.

## OpenAI-compatible backend (Mistral / Cerebras / Ollama / ...)

Set `OPENAI_API_BASE` to point at any OpenAI-compatible Chat Completions endpoint. `AI_TOKEN` is sent as a Bearer token; `MODEL` selects the model. Localhost `http://` is permitted for self-hosted backends; all other URLs must be `https://`.

| Provider | `OPENAI_API_BASE` | Example `MODEL` |

| --- | --- | --- |

| Mistral (Devstral) | `https://api.mistral.ai/v1` | `devstral-small-2505` |

| Cerebras | `https://api.cerebras.ai/v1` | `llama-3.3-70b` |

| Groq | `https://api.groq.com/openai/v1` | `llama-3.3-70b-versatile` |

| Together | `https://api.together.xyz/v1` | `meta-llama/Llama-3.3-70B-Instruct-Turbo` |

| Ollama (self-hosted) | `http://localhost:11434/v1` | `devstral-small-2` |

Caller workflow examples

Mistral Devstral (cloud):

```yaml

with:

  AI_TOKEN: ${{ secrets.MISTRAL_API_KEY }}

  MODEL: devstral-small-2505

  OPENAI_API_BASE: https://api.mistral.ai/v1

```

Cerebras (fast inference):

```yaml

with:

  AI_TOKEN: ${{ secrets.CEREBRAS_API_KEY }}

  MODEL: llama-3.3-70b

  OPENAI_API_BASE: https://api.cerebras.ai/v1

```

Self-hosted Ollama:

```yaml

with:

  AI_TOKEN: ollama-no-auth

  MODEL: devstral-small-2

  OPENAI_API_BASE: http://localhost:11434/v1

```

See [`docs/integrations.md`](docs/integrations.md) (Path B) for cost comparison and caveats.

## License

[Apache-2.0](LICENSE)

ecosyste.ms

Data

Tools

Indexes

Applications

Experiments

Awesome

https://github.com/qte77/gha-issue-triage

Awesome Lists containing this project

README