JayBrief

Personal stock-investing news PWA — themed feed (semiconductors / software tech) plus four LLM-written decision-support briefings per day in Korean.

• Deploy: GitHub Actions assembles and publishes a GitHub Pages artifact.
• No application server: the deployed app is a pure client that reads generated JSON.
• Sister project: jayfit (same PWA + Pages pattern).

How it works

pages.yml
  scheduled refresh -> fetch feeds -> optional Gemini digest
                    -> replace the one-commit runtime-data snapshot
                    -> combine main shell + runtime data
                    -> deploy a GitHub Pages artifact

Schedules are best-effort: GitHub-hosted cron routinely lands late, so the 20-minute feed schedule effectively refreshes about once an hour, and digest editions can start 30-60 minutes after their slot (measured 2026-07-20, median 65 min between feed runs).

The PWA (vanilla HTML/CSS/JS) reads generated JSON with cache: 'no-cache' and renders three tabs: 피드 (article list), 브리핑 (LLM digest), and 주요 이벤트 (watchlist calendar and agenda). Application history stays on main; generated JSON lives on the force-with-lease protected runtime-data branch as one rolling root commit, so recurring feed and digest updates do not inflate main history.

Setup

1. Create a free Gemini API key in Google AI Studio.
2. Add it as the GEMINI_API_KEY repo secret (gh secret set GEMINI_API_KEY).
3. Optional: override the model with a GEMINI_MODEL env/variable (default gemini-3.1-flash-lite).

Without the secret, the feed and event calendar still work; only digest generation is skipped. A failed or invalid Gemini response leaves the last good digest untouched.

Development

pip install -r requirements.txt
python scripts/fetch_feeds.py          # create/refresh ignored data/feed.json
GEMINI_API_KEY=... python scripts/make_digest.py
python -m http.server 8000             # then open http://localhost:8000

The data/ directory is runtime output and is intentionally ignored except for .gitkeep. Run the fetcher locally before developing views that require feed or digest data.

Conventions and data contracts: see CLAUDE.md.