issuesort

Triage GitHub issues with AI — find duplicates and suggest labels. Zero dependencies.

When a new issue lands, issuesort compares it against the open issues by meaning (embedding similarity) to surface likely duplicates, and asks a model to pick fitting labels from your repository's own label set. It posts the result as a comment, and can apply the labels for you.

Run it as a GitHub Action on every new issue, or invoke the CLI by hand. It uses any OpenAI-compatible API — including a local server — and nothing but the Python standard library.

Example comment

## issuesort triage

**Possible duplicates:**
- #128 App crashes when saving an empty file _(similarity 0.91)_
- #97 Save dialog throws on empty path _(similarity 0.86)_

**Suggested labels:** `bug`, `needs-repro`

Install

pip install issuesort

Requires Python 3.8+.

CLI usage

export OPENAI_API_KEY=sk-...
export GITHUB_TOKEN=ghp_...

# Print a triage for issue #42
issuesort --repo octocat/hello-world --issue 42

# Post it as a comment and apply the suggested labels
issuesort --repo octocat/hello-world --issue 42 --post --apply-labels

Options

GitHub Action

Triage every newly opened issue. Store your provider key as a secret:

name: issuesort
on:
  issues:
    types: [opened]

permissions:
  contents: read
  issues: write

jobs:
  triage:
    runs-on: ubuntu-latest
    steps:
      - uses: Sev7nOfNine/issuesort@v0.1.0
        with:
          api-key: ${{ secrets.OPENAI_API_KEY }}
          apply-labels: "false"

How duplicate detection works

The issue's title and body are embedded and compared against every open issue by cosine similarity. Anything at or above --threshold is reported, best match first. Tune the threshold to taste: higher is stricter.

Privacy note

Issue text is sent to whichever provider you configure. Prefer a self-hosted model via --base-url for private repositories.

Development

pip install -e ".[test]"
python -m pytest

Tests run offline; the network layers accept injectable fakes.

License

MIT — see LICENSE.