ContribGuard

AI contribution triage for open source maintainers. Detect duplicate issues, AI-generated content, low-quality submissions, and submission velocity anomalies — without any LLM API calls.

Quick Start

npx @phoenixaihub/contrib-guard scan --repo facebook/react

Features

• 🔁 Duplicate Detection — TF-IDF + cosine similarity clustering
• 🤖 AI Content Fingerprinting — Stylometric heuristics (sentence uniformity, vocabulary diversity, pattern matching)
• 📊 Quality Scoring — Checks for repro steps, tests, code references, environment info
• 🚨 Velocity Anomaly Detection — Flags burst submission patterns
• 🔗 Cross-Reference Engine — Matches open issues against closed/fixed ones

Install

npm install -g @phoenixaihub/contrib-guard

CLI Usage

# Scan a repo
contrib-guard scan --repo owner/repo

# With GitHub token for higher rate limits
contrib-guard scan --repo owner/repo --token ghp_xxx

# JSON output
contrib-guard scan --repo owner/repo --json

# Limit items fetched
contrib-guard scan --repo owner/repo --limit 50

GitHub Action

name: ContribGuard Triage
on:
  issues:
    types: [opened]
  pull_request:
    types: [opened]

jobs:
  triage:
    runs-on: ubuntu-latest
    permissions:
      issues: write
      pull-requests: write
    steps:
      - uses: phoenix-assistant/contrib-guard@main
        with:
          github-token: ${{ secrets.GITHUB_TOKEN }}
          ai-threshold: '40'
          quality-threshold: '40'

Action Inputs

| Input | Default | Description | |-------|---------|-------------| | github-token | ${{ github.token }} | GitHub API token | | ai-threshold | 40 | AI detection score threshold (0-100) | | quality-threshold | 40 | Quality score threshold (0-100) | | label-duplicates | possible-duplicate | Label for duplicates | | label-ai | ai-generated | Label for AI-flagged items | | label-low-quality | needs-details | Label for low quality items |

How It Works

Duplicate Detection

Uses TF-IDF vectorization of issue/PR text, then computes pairwise cosine similarity. Items with similarity ≥ 0.6 are clustered as duplicates.

AI Fingerprinting

Pure heuristic analysis — no LLM calls:

• Sentence uniformity: AI text has unusually consistent sentence lengths (low coefficient of variation)
• Vocabulary diversity: AI text tends toward lower type-token ratios
• Pattern matching: Detects hedging phrases ("it is worth noting"), excessive numbered lists, overly formal language

Quality Scoring

Checks issues for: reproduction steps, expected/actual behavior, environment info, code snippets, descriptive titles. Checks PRs for: description, issue references, test mentions, reasonable size.

Velocity Detection

Flags users who submit ≥ 5 items within a 60-minute window.

Cross-Referencing

Matches open issues against closed ones using the same TF-IDF similarity engine to surface "already fixed" reports.

Programmatic API

import { triage, fetchItems } from '@phoenixaihub/contrib-guard';

const items = await fetchItems('owner/repo', 'ghp_token');
const report = triage('owner/repo', items);
console.log(report.summary);

License

MIT