Skip to main content

​
Text detector selection

Core 0.4.3 adds optional entities selection to text scans across Rust, Python, Node.js, and browser WASM. Only selected detectors run, avoiding the cost of scanning for types the caller does not need. This includes the implementation from PR #47. For example, Python callers can scan for email alone:

​
Compatibility

  • Omit entities to preserve existing detector defaults.
  • Use canonical entity names; empty selections, duplicate names, and unsupported names are rejected as configuration errors.
  • Keep explicit activation settings for optional detectors such as UUID and locale-specific types; selection does not enable them automatically.
  • Apply selection to text scans only. Structured scanning is unchanged.
  • Preserve finding order, source offsets, email source-context handling, and the existing capabilities contract.
MCP must adopt this release and pass its selected types into Core to benefit from detector selection. Its existing response filtering alone does not avoid unselected detector work.

​
Verification and publication

Release checks cover version alignment, Rust formatting, Clippy, workspace tests, installed Python and Node packages, and browser WASM. Selection regressions cover supported types, invalid configurations, optional detector activation, source formats, Unicode offsets, and transformations. All runtime packages and lockfiles are aligned at 0.4.3. Runtime publication uses rust-v0.4.3, python-v0.4.3, node-v0.4.3, and wasm-v0.4.3 on the same release commit. Candidate checks do not establish registry availability; publication and clean registry installation must be verified separately.