Available AI skills · SEC EDGAR
Primary Source Research Pack
Make your AI cite the filing every time it gives you a number.
See it in action
Real outputs from the pack's test run on public companies, using SEC filings retrieved on 18 October 2026. The companies are test cases, not recommendations.
What you download. Eight skills, each with its instructions, the shared EDGAR tool and reference files. Example output 
earnings-release-digest on DKS: the guidance table compares two releases, every figure sourced. Example output 
insider-activity-check on BORR: three “purchases” flagged as option exercises. Example output 
annual-report-red-flags on DKS: boilerplate filtered out, no strong flags on a clean company. Example output 
edgar.py lookup. The shared tool's JSON output from the SEC's free API. No API key. Example output
You ask an AI about a stock. It tells you insiders are buying, revenue grew 30% and the company is “financially healthy.” Some of that is real, some of it is made up, and it's hard to tell which.
The Primary Source Research Pack is eight ready-to-install AI skills that work only from SEC EDGAR, the official record of what US-listed companies file. Every number comes with the form, the date and a link to the filing. When the filings don't answer the question, your AI says so.
It's built and tested as a Claude Code plugin. The skills use the open Agent Skills format, so other agents that read it (such as Codex, Cursor or Gemini CLI) may load them too, but we haven't tested those. The included command-line tool also runs on its own, with no AI required.
What you get
| Skill | What it does |
|---|---|
sec-filing-finder | Finds the exact 10-K, 10-Q, 8-K, proxy or prospectus page that answers your question, and quotes it. |
annual-report-red-flags | Going-concern doubt, material weaknesses, auditor changes, late filings, delisting notices and debt acceleration. Boilerplate is filtered so a clean company screens clean. |
earnings-release-digest | The company's own earnings release turned into one page, with guidance changes found by comparing it with the previous quarter. |
insider-activity-check | Separates real open-market buying from grants, tax withholding and option exercises recorded as “purchases”. |
dilution-tracker | Share count over time, stock-based pay, shelf registrations, at-the-market programmes and offerings. |
fact-check-two-sources | Paste a claim from X, Reddit, a newsletter or another AI and get Confirmed, Partly true or Not supported, with evidence. |
bull-bear-memo | A balanced write-up with a pre-mortem, built only from filings. |
filing-event-timeline | What was filed and when, plus the next likely earnings window estimated from past filing dates. |
Plus a dependency-free Python command-line tool, a quick-start guide, worked examples on real filings and a test suite.
Why it's different
- Numbers come with receipts. An accession number and link on every figure. Calculated numbers are labelled derived and show their formula.
- It catches traps that generic AI and many screeners miss. In our test run on Borr Drilling, about $3.9M of Form 4 “purchases” were actually option exercises. The pack flags them; summing every purchase code doesn't.
- Foreign filers are handled. 20-F and 6-K companies are detected, so your AI doesn't go looking for 10-Ks that don't exist.
- Free data, no subscription. It uses the SEC's own public APIs, with no API key and no monthly data bill.
A real example
From the examples folder included with the pack:
Claim: “Dick's insiders have been buying.”
Confirmed. 9 purchase transactions by 7 insiders between Aug 26 and Sep 30, 2026, about $5.59M in totalderived, from 8 Form 4 filings, each linked.
Example line: EVP & CFO bought 7,707 shares at $129.75 on 2026-09-22.[1]
Data retrieved 2026-10-18 from SEC EDGAR. Shown to demonstrate the format, not as a pick.
Who it's for
Self-directed investors, people who use AI for stock research, newsletter writers who need to cite sources, and builders of trading or research bots who want a primary-source data layer.
Who it's not for
If you want buy or sell signals, price targets or chart analysis, this isn't it. EDGAR has no price data, and the pack never gives advice.
How it's tested
Version 1.0 passes 18 automated tests (13 offline and 5 against live EDGAR data), installs and loads as a Claude Code plugin, and was run end to end on three companies chosen to stress different cases: a large US company, a foreign filer and a distressed small company. Testing found seven bugs, all fixed before release.
Questions
Do I need to know how to code?
No. You set one line (your name and email, which the SEC asks every user to provide) and install the skills. Copy-and-paste instructions are included.
Which AI tools does it work with?
Claude Code, where it's tested as a plugin. The Claude apps with code execution turned on and other Agent Skills tools, such as Codex, Cursor and Gemini CLI, should be able to load the skills, but we haven't tested them. Without a skills-capable AI, you can run the included command-line tool and paste its output into any chat.
Does it cost anything to run?
The SEC data is free and needs no API key. Your AI tool's usage costs are whatever your plan already charges.
Does it cover non-US stocks?
Only companies that file with the SEC, including foreign companies listed in the US (20-F and 6-K filers). It doesn't cover companies listed only in London, Toronto and so on.
Can it get earnings-call transcripts?
No. They aren't on EDGAR. The pack uses the company's filed press release (Exhibit 99.1) and 10-Q instead.
Will it tell me what to buy?
No. It's built not to. Each skill refuses buy, sell or hold calls and ends with a list of its sources.
Is the data real-time?
Filings usually appear on EDGAR within minutes of being accepted. The pack caches results for 12 hours by default, and you can change that.
What system do I need?
Python 3.9 or newer. The command-line tool uses only Python's standard library, so there's nothing else to install.
Can I get a refund?
Yes. If it isn't right for you, email us within 14 days of purchase for a full refund, no questions asked. See the refund policy.
Disclaimer: research and education tool only, not investment advice. The seller is not a registered investment adviser or broker-dealer. No claim is made that using this product will improve investment results. Data comes from public SEC filings as filed and may be amended. Always verify figures in the linked filing. SEC EDGAR is operated by the U.S. Securities and Exchange Commission, which does not endorse this product.