Decoy redact

Manually masking sensitive data in documents is error-prone and slow. This skill automates the redaction of PDFs and images using the decoy CLI, ensuring secrets are covered by pixelated phrases rather than simple black boxes.

What it does

• Pattern matching identifies sensitive strings like IBANs, emails, or names using JavaScript regex.
• Visual masking replaces secrets with a mosaic generated from decoy phrases to prevent reconstruction.
• Automated verification confirms that redacted files contain no residual text layers, metadata, or readable secrets.
• Local processing performs all operations on the host machine without uploading files to external servers.
• Visual blending samples surrounding ink and paper colors so redactions integrate naturally with the document.

How it works

1. Find involves scanning the PDF text layer to identify specific strings and their coordinates.
2. Redact applies regex patterns or explicit pixel coordinates to generate a flattened, masked version of the file.
3. Verify compares the output against the original to prove no sensitive data remains and only intended areas were altered.
4. Report generates a summary of covered pages, boxes, and the final output path.

Frameworks & tools

This skill utilizes the decoy CLI tool via Node.js/npm and supports integration through MCP. It relies on pdf.js for text layer analysis and standard JavaScript regular expressions for pattern matching.

Why this beats prompting it yourself

Standard LLM prompting cannot reliably modify binary file structures like PDFs or guarantee the removal of hidden text layers. This skill uses a specialized CLI to ensure the underlying data is physically removed and replaced, providing a verifiable audit trail that simple image generation lacks.

Use cases

• Redact bank account numbers and IBANs from financial statements.
• Anonymize PII like names and email addresses in legal discovery documents.
• Mask sensitive areas in screenshots or scanned images using coordinate boxes.
• Sanitize internal reports for public distribution while maintaining visual layout.

Known limitations

Scanned PDFs and images require explicit coordinate boxes as they lack a text layer. The size of the redaction box remains proportional to the hidden text, potentially revealing secret length.