dsh-knowledge-graph: An Ecosystem and Turn-Trace Graph for DeepSeek Harness
rickdeng005/dsh-knowledge-graph
A DSH plugin that renders installed plugins, skills and tools plus their dependencies as a force-directed graph, and visualizes each conversation turn's tool calls and reasoning, with usage stats, health checks and report export.
dsh-knowledge-graph is a visualization plugin for DeepSeek Harness that draws every installed plugin, skill and tool, together with their dependency and bundle-containment relations, into one interactive force-directed graph. It targets the usual pain of a bloated environment, invisible broken dependencies, an opaque agent whose per-turn behavior is guesswork, and entries that sit unused. Beyond the global graph it offers a list view for precise lookup, per-turn reasoning replay, usage statistics and health diagnosis, plus one-click Markdown report export, with stats and history persisted locally across restarts.
How to Install
dsh plugin --profile web add github:rickdeng005/dsh-knowledge-graph- Category
- Data & Knowledge
- Platform
- DSH Plugin
- Author
- rickdeng005
- Distribution
- Plugin
dsh-knowledge-graph Key Features
dsh-knowledge-graph Repo Summary
What Does It Do?
DSH Knowledge Graph is a DSH plugin for DeepSeek Harness that turns your installed plugins, skills and tools into an interactive visual graph. It solves the problem of not knowing what is installed, how items depend on each other, what the agent actually did each turn, and which plugins are dead weight. Its core capabilities include a force-directed relationship graph, list views, per-turn reasoning traces, usage statistics, health diagnosis and one-click report export. It is maintained by RickDeng005 under the MIT license and written in JavaScript.
Core Features
- Force-directed graph: plugins, skills and tools appear as nodes with a smooth physics layout; dependency relations are drawn as edges, while bundle containment is grouped into convex-hull bubbles so you can see which items belong together.
- List view: three tabs for plugins, skills and tools with name search and layer filtering, suited to precise lookup of a single entry.
- Per-turn thinking: selecting a conversation turn highlights the tools it called with sequence badges and call arrows, and draws reasoning and reply as bubbles; click a bubble for the full text, and clicking a tool node highlights its related bubbles.
- Usage stats and health diagnosis: each node shows call count, last-used time and recent turns it appeared in, while failed and disabled plugins are marked with colored outer rings.
- Export and local persistence: one-click Markdown report covering overview, problem plugins, dormant list and usage ranking, with stats and history stored locally so they survive restarts.
How to Use This Plugin?
After enabling it in DSH, open the graph view to get the whole picture: nodes stand for plugins, skills and tools, edges for dependencies, and bubbles for bundle containment. Switch to the list view and use search or layer filtering when you need to locate a specific item; if you suspect a broken dependency, open the full-dependency display to inspect relation edges or a plugin's dependency bubble. To review a conversation, use the turn selector to replay any turn, and click a tool node to highlight its related reasoning bubbles; to clean up your environment, use the dormant list or export a report to find plugins and tools that were never used.
How to Troubleshoot This Plugin?
Plugins that failed to load are marked with a red outer ring in the graph, while disabled ones are marked grey, so you can quickly locate broken entries. If a feature misbehaves because of a broken dependency chain, open the full-dependency display to inspect relation edges or the plugin's dependency bubble, confirm the missing dependency, and repair it according to the official instructions.
This page is an independent rewrite of the plugin's official README — for authoritative documentation and the latest changes, refer to the source: rickdeng005/dsh-knowledge-graph. The plugin is third-party code that runs on your machine once installed; inclusion does not imply endorsement — please review the source before installing.
