dsh-graphify-plugin

When a model joins a large codebase, it rarely has a global picture of architecture: which modules depend on which, which clusters form the structural core, how far one pull request's blast radius reaches. These answers live in code but resist a simple natural-language question. dsh-graphify-plugin bridges DeepSeek Harness to the local knowledge-graph engine Graphify, giving the model a set of graph-query tools (architecture Q&A, dependency paths, node neighborhoods, community detection, god-node identification, shortest-path analysis) so it reads structural semantics directly instead of piecing together hints through repeated searches.

The plugin owns the Graphify MCP subprocess end-to-end: it spawns the service the first time an agent calls a graph tool and reaps it when the DSH session ends, with no manual daemon. Because the graph path is resolved from the calling agent's session directory, a single interactive DSH process can serve multiple workspaces without cross-contamination. Interactive adapters expose a /graphify slash command for manual graph builds, and the DSH Web browser companion persists both successful output and failures into the session event stream so they survive a page reload. On the review side, list_prs, get_pr_impact, and triage_prs let the model quantify a pull request's impact before it is merged.

This plugin is built for developers and teams working inside DeepSeek Harness who want their agent to reason about code structure at the architecture level: when you need the model to understand module boundaries, trace cross-file dependency chains, spot architectural hotspots, or gauge a PR's blast radius before merge, dsh-graphify-plugin hands those graph capabilities to the model as ready-made tools and removes the need to hunt through design docs by hand.