SEC EDGAR Filing Research

Why PDF instead of XBRL? SEC filings contain inline XBRL (HTML with embedded XML tags) which creates noise for LLMs. Additionally, ~34% of filings have XBRL tagging errors, companies use non-standard tags, and narrative content (MD&A, Risk Factors) isn't captured in XBRL. PDF + file_perception reads what humans see.

Rate limit: 10 requests/second. All requests require header: User-Agent: ShortcutAgent/1.0 (support@shortcut.ai)

Workflow Overview

1. •Find company → Get CIK from ticker/name
2. •Fetch metadata → Get filing list with dates from SEC API
3. •VERIFY "LATEST" → Sort by date, confirm against today's date (CRITICAL!)
4. •Convert to PDF → Use sec_to_pdf.py script (Playwright + Chrome)
5. •Analyze → Use file_perception on the PDF

API Reference

All requests require header: User-Agent: ShortcutAgent/1.0 (support@shortcut.ai)

Rate limit: 10 requests/second

Step 1: Find Company CIK

Fetch company tickers JSON and search for the ticker:

GET https://www.sec.gov/files/company_tickers.json

Returns: {"0": {"cik_str": 320193, "ticker": "AAPL", "title": "Apple Inc."}, ...}

Zero-pad CIK to 10 digits for the next step (e.g., 320193 → 0000320193).

Step 2: Fetch Filing Metadata

GET https://data.sec.gov/submissions/CIK{cik_10digit}.json

Response contains filings.recent with parallel arrays: form, filingDate, reportDate, accessionNumber, primaryDocument. Index i across all arrays gives one filing's info.

Step 3: VERIFY "Latest" Filing (CRITICAL)

NEVER assume the first result is the latest. The API doesn't guarantee sort order.

Verification Protocol

1. •

   Check today's date from the system prompt (e.g., "Today's date: 2026-01-06")

2. •

   Collect ALL filings of the requested type from the parallel arrays

3. •

   Sort by filingDate descending to find the truly latest

4. •

   Sanity check against expected filing schedule:

   Filing	Expected Timing
   10-K	60-90 days after fiscal year end
   10-Q	40-45 days after quarter end
   8-K	Within 4 business days of event
   DEF 14A	~120 days before annual meeting

5. •

   Report exact dates to user:

   "Found Tesla's latest 10-K filed on 2026-01-29 for fiscal year ending 2025-12-31. Is this the one you need?"

Red Flags

• •If today is January 2026 but the "latest" 10-K shows FY2024 → something's wrong
• •If reportDate is more than a year old → re-verify or warn user
• •Multiple filings same day → check for amendments (10-K/A vs 10-K)

Step 4: Convert to PDF

Build URL from metadata:

https://www.sec.gov/Archives/edgar/data/{cik}/{accession_no_dashes}/{primaryDocument}

• •Remove dashes from accessionNumber for URL path
• •Use unpadded CIK (no leading zeros)

Example:

• •CIK: 320193
• •Accession: 0000320193-25-000123 → 000032019325000123
• •Document: aapl-20250928.htm
• •URL: https://www.sec.gov/Archives/edgar/data/320193/000032019325000123/aapl-20250928.htm

Use the sec_to_pdf.py script to convert directly from URL (renders images properly):

python /skills/default/sec-edgar/sec_to_pdf.py "<filing_url>" "/workspace/sec/{TICKER}/{form}_{report_date}.pdf"

File path convention: Always save SEC filings to /workspace/sec/{TICKER}/{form}_{report_date}.pdf

• •Example: /workspace/sec/AAPL/10-K_2024-09-28.pdf
• •Example: /workspace/sec/TSLA/10-Q_2024-06-30.pdf
• •This keeps filings organized by company and makes them easy to find later

Why Chrome? Two reasons: (1) SEC.gov blocks headless Chromium with 403 errors - real Chrome bypasses this. (2) Downloading HTML locally breaks relative image paths; Chrome navigating directly to the URL keeps images loading from SEC's servers.

Fallback: If the script fails, you can use httpx to download locally and weasyprint amongst other tools and packages to convert the html to PDF. If you download HTML first and use filename=, relative image paths may break and images may not get rendered. You will have to check and fix this.

Step 5: Analyze with file_perception

Use file_perception on the PDF to extract information.