redact-pii-skilllisted

# PII Detector Finds personally identifiable information in a block of text and produces a structured report. Combines two complementary detection methods so nothing gets missed: 1. **Pattern-detectable PII** (has a fixed, verifiable shape — SSNs, emails, credit cards, IPs, IBANs, VINs, etc.) — run through `scripts/scan_pii.py`, which uses regex + checksum validation (e.g. Luhn check for card numbers) for high-precision, deterministic matches. 2. **Context-detectable PII** (no fixed shape — names, race/ethnicity, diagnoses, street addresses, employment history, behavioral profiles, etc.) — read directly by Claude, checked against the full category checklist in `references/pii_taxonomy.md`. Relying on regex alone misses free-text PII; relying on the model alone misses easily-verified structured identifiers and can hallucinate on ambiguous digit strings. Always do both passes. ## Workflow 1. **Get the text.** If it's in a file, read the file. If it's pasted inline, use it directly. 2. **Run the pattern scan:** ```bash python3 scripts/scan_pii.py <path-to-file> # or, for inline text: python3 scripts/scan_pii.py --text "the text to scan" ``` This returns a JSON array of findings with `category`, `subtype`, a masked `match` value, character offsets, and a `confidence` level (`high`/`medium`/`low`). Treat `low` confidence hits as candidates to double check against context, not confirmed PII — e.g. a bare 5-digit number flagged as a possi