NLP++ Weekly: help system, LLM prompts, load-time visibility, and a busy release week

General
1

Big week — lots of small, practical improvements that add up. Let me walk through the themes.

New in-editor help system + LLM prompt library

I landed a real help panel inside the VS Code extension (3.2.0), with:

  • A built-in LLM Prompts view — browse and use prompt templates without leaving the editor, now with hover descriptions so you know what each one does before you click it.
  • Announcements support and a “Show Latest” shortcut, so I can push news to you without you having to dig through changelogs.
  • 91 new NLP++ function snippets pulled straight from the help docs — autocomplete should feel a lot more useful now.
  • Fixed LLM Prompts opening as raw text instead of a rendered preview.

I also cleaned up the toolbar — Video and Create Claude Prompt moved into the ... overflow menu so the treeview title bar isn’t so cluttered.

Knowing where your time actually goes

This was the theme of the back half of the week: load-time and analyze-time visibility. If you’ve ever wondered why an analyzer feels slow to start, you should be able to answer that now:

  • The engine reports KB, dictionary, and overall analyzer load times separately.
  • The extension shows lazy-loaded KB/dictionary timing in the analyzer summary, with its own dedicated log file instead of mixing into the general output.
  • Analyze timing summaries got a cleanup pass, and the log now persists on completion instead of vanishing.

Small thing but a real annoyance fixed: the engine updater was re-downloading the engine on every single startup. That’s stopped now.

Dictionary/KB tooling

  • New byte-sort command for full.dict / full.kbb files, right from the KB view.
  • Fixed a bug where unsorted *full.kbb/*full.dict files were being loaded normally instead of being skipped — could cause subtle lookup issues.
  • New Knowledge Base template default.

New NLP++ functions

Engine v3.7.0 adds file-system functions directly callable from NLP++: fileexists, direxists, filesize, deletefile. Useful if your analyzers need to check or manage files as part of a rule set instead of shelling out.

Regression testing

Fixed bless regression so it only warns when goldens already exist, instead of complaining on first run.

Releases

A genuinely heavy release week — 34 releases across the org:

  • nlp-engine: v3.6.3 → v3.6.5 → v3.7.0 (file-system functions, load-time reporting, dictionary skip fix)
  • vscode-nlp: v3.1.30 → v3.2.12 (help system, LLM prompts, snippets, timing UI, byte-sort)
  • py-package-nlpengine: v2.1.1 → v2.1.5 (engine bumps + full cp310–cp313 macOS wheels via cibuildwheel)
  • Windows/Linux/Mac engine bundles all synced up to v3.7.0

If you’re on the Python package, the macOS wheel coverage should make installs a lot smoother across Python versions.

What’s next

Now that load-time reporting is in place, I want to actually chase down the slow spots it’s surfacing — dictionary and KB loading in particular. More on that next week.