XSD-aware streaming XML obfuscator

TypeScript/Node.js streaming obfuscator intended for very large XML documents. It does not load the input XML into memory.

Install

Run without installing globally:

npx xsd-aware-xml-obfuscator \
  --xsd schema.xsd \
  --input input.xml \
  --output output.xml \
  --secret "$XML_OBFUSCATION_SECRET"

Or install the CLI globally:

npm install --global xsd-aware-xml-obfuscator
xsd-xml-obfuscate --help

Develop/build

npm install
npm run build

Run

xsd-xml-obfuscate \
  --xsd "s3000l_3-0_lsaDataset.xsd" \
  --input input.xml \
  --output output.xml \
  --secret "$XML_OBFUSCATION_SECRET"

Add --readable-names to replace values of name elements and attributes with deterministic readable pseudonyms such as Amber Falcon.

Use a long random secret and do not commit it to source control.

Design

• input XML: streaming SAX parser;
• XSD: loaded into an in-memory rule model;
• named complexType and group declarations are expanded where possible;
• deterministic HMAC-based masking;
• xsd:ID, xsd:IDREF, and xsd:IDREFS share one masking domain so references remain consistent;
• enumeration values stay inside their XSD enumeration;
• unknown/unresolved XSD types are preserved, not guessed;
• output is streamed to disk.

The CLI prints counts including changed, preservedBySchema, and unresolvedPreserved.

See SCHEMA_NOTES.md for findings from the supplied files and the missing schema dependencies.

Current limitations

This is still an MVP, not a complete implementation of the XSD 1.1 specification. Arbitrary regex generation for xsd:pattern, XSD 1.1 assertions as transformation rules, substitution groups, wildcards, type alternatives, and every possible schema composition construct are not fully modeled.

For safety, unresolved values are preserved. This avoids producing values that look obfuscated but violate S3000L constraints.

License

MIT