Data & Knowledge DSH plugins

Knowledge plugins help the harness remember and reason over structured material: retrieval over documents, research and deep reading workflows, knowledge graphs, diagram generation, notebooks, and visualization of whatever the agent produced.

The useful signal here is provenance. A knowledge plugin that shows where a claim came from is worth more than one that only produces a confident summary, and the best ones let you trace a sentence back to a source file.

How to choose in this category

Ask where the corpus lives and how a hit is justified. Local Markdown trees and PDFs with page references are easier to audit than opaque vector stores with no source ids. Diagram and architecture tools should emit files you can diff, not only images that die in chat history.

Research pipelines earn trust when every claim can name a file, URL, or page. Visualization is only as good as the data path behind it — a pretty graph over unvalidated extracts is still unvalidated.

What is mature versus thin here

Mature listings ship evidence-shaped output, explicit corpus boundaries, and export formats other tools can read. Thin ones summarize a folder and call it a knowledge base. Provenance labels that the project assigns itself (not third-party certification) should be read as such.

Common mistakes

Mistaking retrieval volume for understanding. Storing secrets inside a “knowledge” corpus the model will re-read. Generating diagrams that cannot be regenerated because the source JSON was never saved.

Install order and boundaries

For knowledge plugins, ask first: where does the corpus come from, is retrieval bounded, and are writes reviewable? A “knowledge” plugin without those two is a larger prompt cache.