AgentWard Sanitize

Detect and redact PII from text files. Supports 15 categories including credit cards, SSNs, emails, API keys, addresses, and more — with zero dependencies.

Security Scan

The skill's code, requirements, and runtime instructions are consistent with a local PII redaction tool that uses only Python and no external services.

Purpose & Capability

Name/description (PII detection & redaction) match the delivered artifacts: a Python3 script implementing regex/Luhn-based detectors and a docs file listing supported categories. Requiring only python3 is proportional.

Instruction Scope

SKILL.md gives clear runtime rules to avoid reading raw input or the entity-map and to prefer --output/preview/json. That is appropriate for minimizing exposure of raw PII, but it relies on the agent actually following those rules. The docs also mention a plain 'sanitize to stdout' usage and state 'PII never reaches stdout' — this is likely because sanitized output replaces values with placeholders, but the mixed wording could confuse non-technical users. Important: the script writes a sidecar entity-map containing raw PII when --output is used; the README explicitly instructs not to read that file.

Install Mechanism

No install spec and only a single Python script included; no downloads or external packages. Lowest-risk install posture.

Credentials

No environment variables, credentials, or external service tokens are requested. The API-key detection patterns include common provider prefixes (OpenAI, GitHub, AWS, Slack) for detection only; they are not used to access those services.

Persistence & Privilege

always is false and the skill does not request elevated or persistent privileges. It does write output and an entity-map file to disk (when --output is used), which is normal for a redaction tool.

Assessment

This appears to be a straightforward, local PII redaction tool. Before installing or running it: 1) Review and keep the script local; it uses only Python stdlib and makes no network calls according to the code and docs. 2) Be aware the tool writes an entity-map sidecar containing raw PII when you use --output; protect or securely delete that file (do not let the agent or other processes read it). 3) Ensure your agent runtime actually follows the SKILL.md rule not to open the raw input or the entity-map — the safety depends on that behavior. 4) Prefer --preview or --json for inspection and run the tool on a sandbox or test data first. 5) If you need absolute assurance, open and scan the full sanitize.py file yourself (the included code appears benign).

Detect and redact personally identifiable information (PII) from text files.

IMPORTANT — PII Safety Rules

• Do NOT read the input file directly. It may contain sensitive PII.
• ALWAYS use --output FILE to write sanitized output to a file.
• Only read the OUTPUT file, never the raw input.
• Only show the user the redacted output, never the raw input.
• --json and --preview are safe — they do NOT print raw PII values to stdout.
• The entity map (raw PII → placeholder mapping) is written to a separate sidecar file (*.entity-map.json) only when --output is used. Do NOT read the entity map file.

What it does

Scans files for PII — credit cards, SSNs, emails, phone numbers, API keys, IP addresses, mailing addresses, dates of birth, passport numbers, driver's license numbers, bank routing numbers, medical license numbers, and insurance member IDs — and replaces each instance with a numbered placeholder like [CREDIT_CARD_1].

Usage

Sanitize a file (RECOMMENDED — always use --output)

python scripts/sanitize.py patient-notes.txt --output clean.txt

Preview mode (detect PII categories/offsets without showing raw values)

python scripts/sanitize.py notes.md --preview

JSON output (safe — no raw PII in stdout)

python scripts/sanitize.py report.txt --json --output clean.txt

Filter to specific categories

python scripts/sanitize.py log.txt --categories ssn,credit_card,email --output clean.txt

Supported PII categories

See references/SUPPORTED_PII.md for the full list with detection methods and false positive mitigation.

Category	Pattern type	Example
credit_card	Luhn-validated 13-19 digits	4111 1111 1111 1111
ssn	3-2-4 digit groups	123-45-6789
cvv	Keyword-anchored 3-4 digits	CVV: 123
expiry_date	Keyword-anchored MM/YY	expiry 01/30
api_key	Provider prefix patterns	sk-abc..., ghp_..., AKIA...
email	Standard email format	[e-mail]
phone	US/intl phone numbers	[phone]
ip_address	IPv4 addresses	192.168.1.100
date_of_birth	Keyword-anchored dates	DOB: 03/15/1985
passport	Keyword-anchored alphanumeric	Passport: AB1234567
drivers_license	Keyword-anchored alphanumeric	DL: D12345678
bank_routing	Keyword-anchored 9 digits	routing: 021000021
address	Street + city/state/zip	742 Evergreen Terrace Dr, Springfield, IL 62704
medical_license	Keyword-anchored license ID	License: CA-MD-8827341
insurance_id	Keyword-anchored member/policy ID	Member ID: BCB-2847193

Security and Privacy

• All processing is local. The script makes zero network calls. No data leaves your machine.
• Zero dependencies. Uses only Python standard library — no third-party packages to audit.
• PII never reaches stdout. The --json and --preview modes strip raw PII values from output. The entity map (containing raw PII to placeholder mappings) is only written to a sidecar file on disk when --output is used.
• Designed for agent safety. The skill instructions above tell the agent to never read the raw input file or the entity map file — only the sanitized output.

Requirements

• Python 3.11+
• No external dependencies (stdlib only)

About

Built by AgentWard — the open-source permission control plane for AI agents.