Weekly Newsletter Digest

A standalone Python pipeline that compiles a week of reading into a single, designed PDF "meta-newsletter" — a navy/teal/gold magazine-style cover, story-by-story summaries, and a clickable bibliography.

It runs entirely on your own machine (cron-friendly, no cloud sandbox), pulling from two sources for a given Monday–Sunday week:

1. Gmail — emails tagged with a News Letter label (read-only Gmail API).
2. Reeder — articles saved to a Reeder read-later shared feed.

Stories from both sources are summarized by Claude via OpenRouter, de-duplicated (Gmail wins), mixed together by date, and rendered to a PDF.

Published as a reference example. Secrets, the feed URL, and the recipient name are placeholders — see SETUP.md.

Sample output

Rendered from synthetic placeholder content — no real names, articles, or subscriptions.

Cover	Inner page

How it works

run.sh (cron entry)
  └─ source mysecrets  →  uv run python -m weekly_digest
        1. fetch_gmail    — News Letter-labeled emails in the Mon–Sun window
        2. fetch_reeder   — read-later items, date-filtered to the same window
        3. dedup          — Gmail wins on overlap
        4. extract        — trafilatura pulls full text for link-only/thin items
        5. summarize      — Claude (OpenRouter) → structured JSON sections + TOC
        6. assemble       — content dict + bibliography + issue number
        7. render         — build_pdf.py → ~/Desktop/weekly_digest_<issue>.pdf

Dependencies are managed by uv from pyproject.toml, so cron never touches a hand-built virtualenv. The PDF layout is intentionally locked (palette, cover coordinates, typography) — it's part of the deliverable, not improvised per run.

Reeder items are filtered by save date (when you added them to the read-later tag), not the article's publish date. Run ./run.sh --since-last-run for an on-demand interim report of everything new since the last interim run; it keeps its own state and never affects the scheduled weekly window.

Layout

weekly-newsletter-digest/
├── run.sh                 # cron entry: sources mysecrets, runs the pipeline via uv
├── gmail_auth.py          # one-time: mints the Gmail read-only refresh token
├── mysecrets.example      # secrets template (copy → mysecrets, gitignored)
├── smoke_test.py          # offline self-test: Reeder + render, no secrets
├── pyproject.toml         # dependencies (managed by uv)
├── SETUP.md               # full setup + cron instructions
└── weekly_digest/
    ├── config.py          # env, Mon–Sun window, issue number
    ├── gmail_source.py    # Gmail API (read-only OAuth)
    ├── reeder_source.py   # date-filtered Reeder feed
    ├── reeder_read_later.py  # zero-dependency Reeder JSON-feed fetcher
    ├── extract.py         # trafilatura full-text extraction
    ├── summarize.py       # Claude via OpenRouter → structured JSON
    ├── assemble.py        # dedup, ordering, bibliography, content dict
    └── build_pdf.py       # the locked ReportLab renderer

Quick start

git clone https://github.com/<you>/weekly-newsletter-digest
cd weekly-newsletter-digest
cp mysecrets.example mysecrets && chmod 600 mysecrets   # then fill it in
./run.sh

Full instructions — Gmail OAuth, OpenRouter, and cron — are in SETUP.md.

Configuration

All settings come from environment variables, sourced from mysecrets:

Variable	Purpose
OPENROUTER_API_KEY	OpenRouter key for summarization
OPENROUTER_MODEL	e.g. anthropic/claude-sonnet-4.6 or ~anthropic/claude-sonnet-latest
GMAIL_CLIENT_ID / GMAIL_CLIENT_SECRET / GMAIL_REFRESH_TOKEN	Read-only Gmail API auth (see gmail_auth.py)
GMAIL_LABEL	Gmail label to compile (default News Letter)
REEDER_FEED_URL	Your Reeder shared feed (the .json suffix is required)
DIGEST_OUTPUT_DIR	Where the PDF lands (default ~/Desktop)
DIGEST_COMPILED_FOR	Recipient name on the cover
DIGEST_WEEK_START / DIGEST_WEEK_END	Optional YYYY-MM-DD override of the default previous-week window

The Reeder fetcher

weekly_digest/reeder_read_later.py is a standalone, standard-library-only fetcher for a Reeder shared JSON Feed (JSON Feed 1.1). It handles the endpoint's 302 redirect to compressed object storage and the JSON Feed 1.0/1.1 author-field differences. The shared feed only holds the last ~50 items, so run often enough that items don't roll off before capture. Note the .json suffix is required on the feed URL.

Cost

The Google/Gmail side is free (well within the Gmail API's free quota; no billing account needed). The only cost is OpenRouter summarization — roughly a few dollars a month, prepaid so you set the ceiling. Pick a cheaper model slug to reduce it.

License

MIT. This project began as a Claude Code skill and was rewritten as a standalone pipeline so it can run unattended via cron.