dsh-csv-parse

In a model-inference pipeline, CSV files are one of the most common structured inputs. Yet real-world CSVs are rarely perfectly tidy—fields contain embedded quotes, commas, or newlines, and encodings can be inconsistent. A single misparsed row can cascade into garbled prompts. dsh-csv-parse exists to handle exactly that: a robust layer of CSV parsing that turns messy tabular data into a structure the model can consume directly.

The entire plugin is a pure Node.js implementation. It makes no network calls, depends on no external services or databases, and runs entirely on the local machine. That makes it safe for offline environments, sandboxed containers, or edge nodes where data must never leave the box. Once registered, the parsing logic is exposed as a session tool—call it, get the result, no extra routing or middleware required.

If you are building inference flows with DeepSeek Harness and your upstream data lands as CSV files while your downstream prompt templates expect structured rows or columns, this plugin fills exactly that gap. It is a good fit for data engineers, automation-script authors, and any developer who wants CSV reading to be reliable and free of runtime dependencies.