Full Disclosure

Full Disclosure pulls 10-K, 10-Q and 8-K filings from SEC EDGAR for a watchlist of companies and has Claude summarize each one: what changed, the biggest risks and the overall tone for investors. A Next.js site publishes the results.

How it fits together

• Pipeline (Python, this folder): edgar.py finds each company's recent filings and keeps the ones worth reading. analysis.py sends the filing text to Claude for structured output and saves the result to Postgres through db.py.
• Schedule: a GitHub Actions workflow (analyze.yml) runs the pipeline every day over watchlist.txt. It analyzes every filing worth summarizing from the past 7 days, and skips filings already in the database. The same workflow can be started for any ticker and lookback window, from the Actions tab or the website's admin-only Analyze button. When it finishes, it tells the website to show the new analyses right away.
• Database: Postgres on Supabase. schema.sql creates the tables. reader.sql creates the read-only login the website uses.
• Website (web/): Next.js and TypeScript, deployed on Vercel. It lists every analysis, and each company has a page with its analyses and its recent EDGAR filings. Search suggests companies from the SEC's full list. Pages are cached for an hour, so new analyses appear without a redeploy.

Running the pipeline locally

python -m venv .venv
.venv/bin/pip install -r requirements.txt
cp .env.example .env   # then fill in the values
.venv/bin/python analysis.py AAPL MSFT
.venv/bin/pytest

analysis.py takes these options:

• --days N: how many days back to look for filings. The default is 7.
• --max N: the most filings to analyze in one run. The default is 15. A run that reaches the limit lists the filings it left, and the next run picks them up.
• --dry-run: list the filings that would be analyzed, without calling Claude or saving anything.

For example, to see what a year of Nvidia's filings would involve:

.venv/bin/python analysis.py --days 365 --dry-run NVDA

Running the website locally

cd web
npm install
cp .env.example .env.local   # then fill in the values
npm run dev

Deploying

1. Run schema.sql, then reader.sql, in the Supabase SQL editor.
2. Add SEC_USER_AGENT, ANTHROPIC_API_KEY and DATABASE_URL (the main login, which can write) as repository secrets in GitHub, under Settings → Secrets and variables → Actions.
3. Import the repo into Vercel with web as the root directory. Set DATABASE_URL to the read-only fd_reader login on the transaction pooler (port 6543), and set SEC_USER_AGENT.
4. Optional, for the admin sign-in: in Supabase, turn off new sign-ups, add the admin's account and add <site>/auth/callback to the redirect URLs. Then set NEXT_PUBLIC_SUPABASE_URL, NEXT_PUBLIC_SUPABASE_PUBLISHABLE_KEY and ADMIN_EMAIL in Vercel and redeploy. Without them the site is read-only.
5. Optional, for the Analyze button: create a fine-grained GitHub token with access to only this repo and "Actions: Read and write", and set it in Vercel as GITHUB_DISPATCH_TOKEN. Set the same random REVALIDATE_SECRET in Vercel and as a GitHub Actions secret.

Design notes

• The CIK lookup is cached, so the SEC's ticker list is downloaded once per run.
• A requests.Session keeps the connection to the SEC open between calls, instead of a new connection and handshake each time.

---

© Rick Bresnahan. All rights reserved. The code is shared for viewing; no license is granted to use, copy or distribute it.