auto-trend：LLM 驱动的 GitHub Trending 日报，自动发布到 GitHub Pages

LLM-powered daily GitHub Trending analysis, auto-published to GitHub Pages.

中文说明

Fork & Deploy

5 minutes to your own trending report.

1. Fork this repo

Fork → uncheck "Copy the master branch only" → Create fork.

2. Enable GitHub Actions

Your repo → Settings → Actions → General → Allow all actions.

3. Set secrets

Settings → Secrets and variables → Actions → New repository secret:

Secret	Description
`LLM_API_KEY`	Your LLM API key (required)
`LLM_BASE_URL`	API endpoint (optional, defaults to OpenAI)
`LLM_MODEL`	Model name (optional, defaults to `gpt-4.1-mini`)
`DAILY_REPO_LIMIT`	Max repos per run (optional, defaults to `20`)
`GITHUB_TOKEN`	GitHub API token (optional, raises rate limit)

4. Enable GitHub Pages

Settings → Pages → Source: Deploy from a branch → Branch: master / /docs.

5. Replace URLs

Replace int2t05 with your GitHub username in these files:

File	Where to update
`docs/_layouts/default.html`	Header GitHub link, footer
`README.md`	Badge URL, Live Reports link

grep -r "int2t05" --include="*.md" --include="*.html" --include="*.js" --include="*.json" .

6. Trigger first run

Actions → Daily Trending Report → Run workflow. Your first report appears at:

https://<your-username>.github.io/auto-trend/

How It Works

GitHub Actions cron (UTC 00:17)
  → Scrape GitHub Trending (httpx + BeautifulSoup)
  → Fetch READMEs concurrently (asyncio)
  → LLM structured analysis per repo (JSON mode)
  → Global trend summary
  → Render Markdown report
  → Generate RSS feed (one item per daily report)
  → Structural verification
  → git commit + push
  → GitHub Pages auto-publish

Each repo is analyzed across 7 dimensions: summary, highlights, core features, use cases, competitive comparison, maturity, and trend signal.

The site itself serves as an RSS source (/feed.xml), providing a subscribable RSS link where each daily report is one item.

The site also supports read marks: toggle read/unread on any report page, and the calendar index dims read dates (state lives in browser localStorage, per device).

Local Dev

pip install -r requirements.txt

export LLM_API_KEY=sk-your-key
export LLM_BASE_URL=https://api.openai.com/v1   # optional
export LLM_MODEL=gpt-4.1-mini                   # optional
export DAILY_REPO_LIMIT=20                      # optional

python scripts/main.py
# Non-CI environments skip git commit/push

Compatible with OpenAI, Anthropic, DeepSeek, or any OpenAI-compatible endpoint.

Environment Variables

Variable	Description	Default
`LLM_API_KEY`	LLM API key	required
`LLM_BASE_URL`	API endpoint	`https://api.openai.com/v1`
`LLM_MODEL`	Model name	`gpt-4.1-mini`
`DAILY_REPO_LIMIT`	Max repos per run	`20`
`GITHUB_TOKEN`	GitHub API token (raises rate limit)	none
`GIT_USER_NAME`	Git commit author name (CI only)	GitHub actor
`GIT_USER_EMAIL`	Git commit author email (CI only)	`<actor>@users.noreply.github.com`

Project Structure

auto-trend/
├── .github/workflows/daily.yml    # Cron trigger
├── scripts/                       # Pipeline
│   ├── main.py                    # Orchestrator
│   ├── config.py                  # Env config
│   ├── fetcher.py                 # Scraper + README fetcher
│   ├── analyzer.py                # LLM analysis (OpenAI SDK)
│   ├── renderer.py                # Markdown report generator
│   ├── rss.py                     # RSS 2.0 feed generator
│   └── verify_report.py           # Post-generation structural check
├── prompts/analysis.md            # LLM system prompt
├── tests/                         # pytest (unit)
├── e2e/                           # Playwright (E2E)
├── docs/                          # GitHub Pages (Jekyll)
│   ├── _layouts/default.html      # Apple-style 3-col layout
│   ├── daily/                     # Generated reports
│   ├── feed.xml                   # Generated RSS feed
│   ├── index.html                 # Report index
│   └── read-state.js              # Read marks (localStorage)
├── requirements.txt
└── package.json                   # E2E dependencies

Testing

# Unit
pip install -r requirements.txt
LLM_API_KEY=sk-test pytest tests/ -v

# E2E
npm install
npx playwright test

Live Reports

int2t05.github.io/auto-trend

License

MIT — see LICENSE.