dsh-contrib-topology

AI tool scheduling control plane + ecosystem plugins for DeepSeek Harness (dsh).

ai-bridge is an umbrella project for making AI coding agents pluggable and observable. It started as a local control plane that orchestrates multiple AI coding tools (Claude Code, OpenClaw, …) via a config-driven tool registry, 7-dimensional Bayesian routing, and a 7-tab web UI — and is now expanding into first-party plugins for dsh ("everything is a plugin").

Packages

Package	Description
@deepseek-ai/dsh-contrib-topology	A dsh plugin that renders a live dependency-topology graph of the running agent: plugin/service nodes, depends-on / provides / contains / disabled edges, and fiber lifecycle phases.

Why topology?

dsh's Cordis kernel makes the runtime a graph — plugins depend on services, services are provided by other plugins, and the whole thing is assembled at boot. But there is no built-in way to see that graph. This plugin exposes a graph() Remote that scans ctx.loader.entries() and produces a TopologySnapshot, then draws it as an SVG in Settings → Plugins → Topology.

Status

• ✅ Integrated into a local deepseek-harness checkout and verified end-to-end (host graph() runtime test, client bundle build, dsh web serves the bundle at /plugins/@deepseek-ai/dsh-contrib-topology/client.js).
• 🟡 Browser visual confirmation pending (local dsh web environment issue under investigation).
• 📝 dsh Discussions intro post drafted — see DSH-DISCUSSIONS-TOPOLOGY-DRAFT.md.

Repo layout

ai-bridge/
├── packages/
│   └── topology/        # @deepseek-ai/dsh-contrib-topology
├── LICENSE
└── README.md

License

MIT