redactlisted

# Redact A multi-pass PII detection and sanitization skill. Strips, masks, or pseudonymizes sensitive identifiers from text, code, documents, and structured data. --- ## Modes Always confirm the mode with the user before proceeding unless it is obvious from context. | Mode | Output | When to use | Legal standard met | |---|---|---|---| | `REDACT` | `[REDACTED]` | Highest privacy — no recovery possible | Exceeds all standards | | `PSEUDONYMIZE` | `[NAME_1]`, `[EMAIL_2]` | Preserve structure, entities distinguishable | Internal use; not de-identification | | `TOKENIZE` | `sha256:a3f2...` (first 8 chars) | Consistent mapping for data analysis | Internal use; not de-identification | | `HIPAA_SAFE_HARBOR` | Remove all 18 PHI identifiers per 45 CFR § 164.514(b)(2) | Healthcare data; producing de-identified data under HIPAA | HIPAA Safe Harbor de-identification | | `GDPR_ANONYMIZE` | k-anonymous output — all quasi-identifiers generalized or suppressed | Producing data outside GDPR scope for research/analytics | GDPR/WP29 anonymization (high bar) | Default to `PSEUDONYMIZE` if the user doesn't specify — it preserves readability. **Critical distinction — pseudonymization vs. anonymization:** - `PSEUDONYMIZE` / `TOKENIZE` produce **pseudonymous** data — still personal data under GDPR Art. 4(1) and HIPAA. Subject to all regulatory obligations. - `HIPAA_SAFE_HARBOR` produces **de-identified** data under HIPAA — outside HIPAA scope if all 18 identifiers removed AND no actual knowle