NoMore Spam

English | 简体中文

AI-powered GitHub Action for detecting spam, low-quality issues, and suspicious pull requests. It can also classify valid issues and pull requests with repository labels.

Features

• Detects spam and meaningless issues or pull requests
• Checks whether an issue is already fully answered by the README or pinned issues
• Requests more information for unclear bug reports
• Validates pull request titles and optionally inspects file changes
• Classifies valid issues and pull requests with configurable labels
• Handles HTTP content-filter errors, Responses API refusals, incomplete output, and failed response states explicitly
• Separates trusted detection instructions from untrusted issue, pull request, and repository content
• Supports GitHub Models and OpenAI-compatible API providers
• Supports English and Simplified Chinese bot responses
• Supports username blacklists

Usage

Create .github/workflows/nomore-spam.yml in your repository:

name: NoMore Spam

on:
  issues:
    types: [opened]
  pull_request_target:
    types: [opened]

permissions:
  contents: read
  issues: write
  pull-requests: write
  models: read

jobs:
  spam-detection:
    runs-on: ubuntu-latest

    steps:
      - name: Detect and close spam
        uses: JohnsonRan/nomore-spam@main
        with:
          github-token: ${{ github.token }}

Bot comments and generated answers use English by default. To use Simplified Chinese:

with:
  github-token: ${{ github.token }}
  language: zh-CN

Inputs

Input	Description	Required	Default
github-token	GitHub token used to read repository content and manage issues or pull requests	Yes	${{ github.token }}
ai-model	AI model name	No	openai/gpt-4o
ai-base-url	OpenAI-compatible API base URL. Omit to use GitHub Models	No	GitHub Models endpoint
ai-api-key	API key for a custom provider. Omit to use the GitHub token	No	Empty
ai-api-type	API interface: chat-completions or responses	No	chat-completions
labels	Comma-separated labels available to AI classification	No	bug,enhancement,question
language	Bot response language: en or zh-CN	No	en
analyze-file-changes	Include limited pull request file changes in analysis	No	true
max-analysis-depth	File analysis depth: light, normal, or deep	No	normal
blacklist	Comma-separated GitHub usernames to close without AI analysis	No	Empty

Full example

name: NoMore Spam

on:
  issues:
    types: [opened]
  pull_request_target:
    types: [opened]

permissions:
  contents: read
  issues: write
  pull-requests: write
  models: read

jobs:
  spam-detection:
    runs-on: ubuntu-latest

    steps:
      - name: Detect and close spam
        uses: JohnsonRan/nomore-spam@main
        with:
          github-token: ${{ github.token }}
          ai-base-url: ${{ secrets.AI_BASE_URL }}
          ai-api-key: ${{ secrets.AI_API_KEY }}
          ai-api-type: responses
          ai-model: ${{ secrets.AI_MODEL }}
          labels: 'bug,enhancement,question'
          language: en
          analyze-file-changes: 'true'
          max-analysis-depth: normal
          blacklist: ${{ secrets.BLACKLIST }}

ai-base-url must be the provider's API base URL, not a full endpoint path. Use ai-api-type: responses for /responses; the default chat-completions uses /chat/completions. Responses requests set store: false so issue and pull request content is not retained for response state.

Detection flow

Issues

1. Close blacklisted users immediately.
2. Detect obvious spam or meaningless content.
3. Check whether the README or pinned issues completely answer the issue.
4. Classify valid issues using the configured labels.
5. Run additional quality checks for bug-like classifications.
6. Ask for missing information or close basic usage questions when applicable.

When an AI provider returns an explicit content-policy rejection such as content_filter, ResponsibleAIPolicyViolation, or a Responses API refusal, NoMore Spam adds a neutral explanation and closes the issue without locking it. Incomplete output caused by token limits, failed response states, and unrelated HTTP 400 responses remain normal action failures.

Pull requests

1. Close blacklisted users immediately.
2. Optionally collect a limited summary of changed files.
3. Detect spam or meaningless content.
4. Validate the pull request title.
5. Check pull request quality and malicious or trivial changes.
6. Classify valid pull requests using the configured labels.

Languages

language controls bot comments and AI-generated README answers:

• en — English, default
• zh-CN — Simplified Chinese

Unsupported values fall back to English. Detection prompts and machine-readable decisions such as SPAM, VALID, and COVERED remain in English for stable parsing.

Permissions

permissions:
  contents: read
  issues: write
  pull-requests: write
  models: read

• contents: read reads the repository README.
• issues: write comments on, labels, closes, and locks issues.
• pull-requests: write comments on and closes pull requests.
• models: read accesses GitHub Models when no custom provider is configured.

License

MIT