dsh-codegraph

In a large codebase, questions like how is this module organized, who calls whom, and why was it designed this way often produce a long, unstructured wall of chat text that is hard to scan and even harder to trace back to evidence.

dsh-codegraph turns that kind of question into a structured architecture scene. The main agent delegates read-only investigation to a continuable child agent; the child returns a validated scene JSON that the plugin renders in better-sidebar with fixed React components and deterministic layout. Node names, relationships, and one-line takeaways stay always visible, while scene notes, fuller explanations, evidence paths, and line ranges expand on demand. The sidebar acts as a read-only viewer, neither a second chat window nor a free-form diagram editor.

It is well suited for developers who use DeepSeek Harness to understand unfamiliar code, review architecture, or onboard into a new repository. You ask one focused question and get a layered, evidence-backed architecture panorama in the sidebar, instead of hunting through scattered chat fragments.