detect_pii

Scan columns for personal data and report likely matches with a confidence score. Read-only. Each column is scored on two clues: its header (email, name, gender, country, birthdate, ip, ...) and its values (email, phone, IP, credit card, IBAN, SSN, date, postal-code shapes), so header-only fields like names or country are found while plain number columns are not.

When to use

• Finding sensitive columns before sharing or exporting a file.
• Driving an anonymisation step: the response includes suggested rules.

Input schema

Parameter	Type	Required?	Default	Description
path	string	no*	(no default)	Path to the file (omit when open_tab is set)
open_tab	string	no	(no default)	Operate on an open GUI tab (@active or tab name)
table	string	no	(no default)	Specific table for multi-table sources
sample_rows	integer	no	500	Rows sampled per column for pattern matching

* path or open_tab is required.

Response shape

{
  "findings": [
    { "column": "email", "kind": "email", "confidence": 1.0,
      "by_name": true, "value_match": 1.0 }
  ],
  "suggested_rules": [
    { "columns": [1], "strategy": { "type": "hash", … }, "new_column": null }
  ]
}

kind is one of email, phone, ip_address, credit_card, iban, ssn, name, gender, country, birth_date, postal_code, address. confidence (0..1, reported when >= 0.5): value_match >= 0.6 -> value_match (+0.2 if by_name); else if by_name -> 0.6 + 0.4 * value_match; else value_match. suggested_rules is ready to pass to the anonymize tool (all detected columns default to a full hash).

Example call

{
  "name": "detect_pii",
  "arguments": { "path": "/data/customers.csv" }
}

See also

• anonymize: apply the suggested (or custom) masking rules.