Sec Edgar

sec-edgar

Response language: match user input.

When to use

Trigger phrases:

• 8-K / 10-K / 10-Q / Form 4 / S-1 / proxy / DEF 14A
• 美股公告 / SEC filing / EDGAR
• insider trading / 内部人交易 / 高管交易
• 财报原文 / 招股书 / 风险因素 / MD&A

Not for: 13F holdings analysis (this skill lists 13F filings but does not parse them into holdings — deferred to a future skill).

Workflow

1. List filings: filings.py <TICKER> (optionally --type 8-K).
2. Read filing text: pass primary_doc_url from the list output to filing_text.py. Use --max-chars to cap; --section item1a for risk factors, --section mda (Item 7) for MD&A.
3. Insider trades: insider.py <TICKER> parses Form 4 XML for the past --days window.

Environment auto-loaded; SEC_USER_AGENT is mandatory.

CLI examples

bashCopy code
# Latest 5 NVDA 8-Ks
python3 .claude/skills/sec-edgar/scripts/filings.py NVDA --type 8-K --limit 5

# All recent NVDA filings
python3 .claude/skills/sec-edgar/scripts/filings.py NVDA --limit 20

# Read the most recent 8-K (URL from filings.py)
python3 .claude/skills/sec-edgar/scripts/filing_text.py \
  "https://www.sec.gov/Archives/edgar/data/1045810/000104581026000051/nvda-20260520.htm" \
  --max-chars 20000

# Pull Item 1A (Risk Factors) from a 10-K
python3 .claude/skills/sec-edgar/scripts/filing_text.py "<10-K URL>" --section item1a

# Save full text to disk, get metadata only
python3 .claude/skills/sec-edgar/scripts/filing_text.py "<URL>" --save-raw /tmp/nvda-10k.txt

# Insider transactions, past 90 days
python3 .claude/skills/sec-edgar/scripts/insider.py NVDA --days 90

# Insider + amendments
python3 .claude/skills/sec-edgar/scripts/insider.py NVDA --include-amendments

Section keys (10-K)

meta.confidence returned: high (clean heading match), medium (short slice, may be incomplete), low (heuristic fallback — full text returned with warning).

Output shapes

filings.py:

jsonCopy code
{
  "data": [
    {
      "accession": "0001045810-26-000051",
      "cik": "0001045810",
      "form": "8-K",
      "filed_date": "2026-05-20",
      "primary_doc_url": "https://www.sec.gov/Archives/edgar/data/1045810/...htm",
      "primary_doc_name": "nvda-20260520.htm",
      "description": "8-K",
      "size": 637530,
      "is_xbrl": true
    }
  ],
  "meta": { "cik": "0001045810", "name": "NVIDIA CORP", "count_returned": 1 },
  "ok": true
}

insider.py:

jsonCopy code
{
  "data": [
    {
      "accession": "...",
      "form": "4",
      "filed_date": "2026-05-15",
      "reporter": "JEN-HSUN HUANG",
      "roles": ["officer:CEO", "director"],
      "txn_date": "2026-05-13",
      "security_title": "Common Stock",
      "code": "S",
      "shares": 240000,
      "price": 412.5,
      "acquire_or_dispose": "D",
      "post_holdings": 78000000,
      "ownership_kind": "D",
      "derivative": false,
      "footnote_ids": ["F1"],
      "footnotes_text": ["Sale pursuant to 10b5-1 plan adopted ..."]
    }
  ],
  "meta": { "cik": "...", "name": "...", "filings_scanned": 12, "txns_parsed": 24 },
  "ok": true
}

Error handling

Known limitations

• insider.py only scans the recent window of submissions (typically last ~1000 filings); deep history requires submissions/CIK*-N.json paging, not implemented.
• filing_text.py uses an HTML-only extractor — XBRL inline tables are stripped to text, not parsed into structured rows.
• Section slicing is regex-heuristic; trust meta.confidence.
• 10 req/s throttle applied globally across all *.sec.gov calls.

Related skills

• longbridge-financial-report for normalised income / balance / cash flow.
• longbridge-news for curated equity news (faster than reading 8-Ks).
• quiver for congressional trades (distinct from corporate insiders).