privacygatelisted

# Data Anonymisation & AI-Safe Processing Turns a document containing personal/sensitive/confidential information into two outputs: a **restricted mapping register** (real values, kept private) and an **AI-safe working copy** (fictitious placeholders only, safe to analyse, summarise, or archive). ## When to actually run this Only proceed if the user has given you specific content to review (pasted text, an uploaded file, or a named document already in context) AND asked for anonymisation/redaction/de-identification. If they just mention privacy in passing, don't invoke this — ask what they want processed instead. ## Workflow 1. **Read the taxonomy.** Load `references/taxonomy.md` for the full identifier categories (direct, indirect, employment, health, financial, technical, special-category). Use it as a checklist, not something to recite back to the user. 2. **Inventory.** Go through the source content once and list every item that matches the taxonomy. Don't skip low-confidence items — flag them as low confidence rather than omitting them. 3. **Re-identification check.** Look for combinations that identify someone even without a direct identifier (e.g. job title + org + city). Rate each Low/Medium/High/Critical with a one-line reason. 4. **Classify.** For each inventoried item assign: security classification (Public/Internal/ Confidential/Restricted), privacy category, confidence (Low/Medium/High), and recommended action (Remove/Generalise/Pseudony