privacy-mask – skill Lập trình cho agent AI

Privacy Mask

Detect and mask sensitive information in images locally before they leave your machine.

Prerequisites

This skill requires the privacy-mask CLI to be pre-installed on the system.

If unavailable, report that prerequisite. The commands below illustrate package names, not authorization to install: select a reviewed version, inspect dependencies and obtain the requested installation scope before changing the environment. Confirm installed help before use:

1. Install via pip: pip install privacy-mask

2. Ensure Tesseract OCR is installed: brew install tesseract (macOS) or apt install tesseract-ocr (Linux)

3. Verify installation: privacy-mask --version

4. (Optional) Install NER support: pip install privacy-mask[ner]

When to use

• User sends a screenshot or image file (.png, .jpg, .jpeg, .bmp, .tiff) that may contain private data
• User mentions privacy, masking, redacting, or anonymizing
• You need to analyze an image but want to redact sensitive info first
• IF the user shares a screenshot for debugging, THEN run privacy-mask mask <path> --dry-run first to check for PII
• IF detections are found, THEN mask the image before proceeding with analysis

Usage

Mask an image:

privacy-mask mask /path/to/screenshot.png
privacy-mask mask /path/to/screenshot.png --in-place
privacy-mask mask /path/to/screenshot.png --dry-run   # detect only, no masking
privacy-mask mask /path/to/screenshot.png --detection-engine regex  # regex only, skip NER
privacy-mask mask /path/to/screenshot.png --config /path/to/custom-config.json

Output is JSON:

{
  "status": "success",
  "detections": [{"label": "PHONE_CN", "text": "***", "bbox": [10, 20, 100, 30]}],
  "summary": "Masked 1 regions: 1 PHONE_CN"
}

Example workflow

1. User provides a screenshot: ~/Desktop/error-screenshot.png

2. Run detection: privacy-mask mask ~/Desktop/error-screenshot.png --dry-run

3. IF detections found, mask the image: privacy-mask mask ~/Desktop/error-screenshot.png

4. The masked output is saved as ~/Desktop/error-screenshot_masked.png

5. Use the masked image for further analysis

What it detects

• IDs: Chinese ID card, passport, HK/TW ID, US SSN, UK NINO, Canadian SIN, Indian Aadhaar/PAN, Korean RRN, Singapore NRIC, Malaysian IC
• Phone: Chinese mobile/landline, US phone, international (+prefix)
• Financial: Bank card, Amex, IBAN, SWIFT/BIC
• Developer keys: AWS, GitHub, Slack, Google, Stripe tokens, JWT, connection strings, API keys, SSH/PEM keys
• Crypto: Bitcoin, Ethereum wallet addresses
• Other: Email, birthday, IP/IPv6, MAC, UUID, license plate, MRZ, URL auth tokens
• NER (optional): Person names, street addresses, organizations, dates of birth, medical conditions

Constraints

• Do NOT send unmasked images to any external API or cloud service
• Do NOT skip masking when detections are found — always mask before sharing
• Do NOT modify the original image unless --in-place is explicitly requested
• Avoid running on very large images (>10MB) without warning the user about processing time

Anti-patterns

• Don't assume images are safe — always run detection even if the image "looks clean"
• Don't use --in-place by default — preserve the original unless the user asks otherwise
• Don't ignore dry-run results — if --dry-run finds PII, the image must be masked before use
• Don't hardcode config paths — inspect the installed CLI default or let the user specify --config

Important

• Verify the installed version and selected detection engine before using sensitive files; do not infer offline behavior from the tool name. Optional models may require downloads.
• OCR and pattern matching can miss secrets, handwriting, small text, QR codes or unusual formats. Inspect the exported image manually before sharing; zero detections does not establish anonymization.
• Treat dry-run output as sensitive: detection logs may include recognized text. Do not publish them or send them to external services unredacted.
• No config.json or CLI is bundled with this AAS entrypoint. Locate defaults in the reviewed CLI installation or pass a user-supplied --config path.

Limitations

• Use this skill only when the task clearly matches its upstream source and local project context.
• Verify commands, generated code, dependencies, credentials, and external service behavior before applying changes.
• Do not treat examples as a substitute for environment-specific tests, security review, or user approval for destructive or costly actions.