creativedswork/dsh-uni-editor

Brings existing editors into DSH agent chat as one entry point, keeping each editor's own UI, data model and domain tools.

Answers the question of what an editor is inside an agent chat. Instead of reimplementing each editor as a chat widget, it brings existing editors in — they keep their own UI, data model and domain tools, while DSH supplies one entry point, the current session context, and a verifiable handoff between human edits and the agent. That handoff is the design goal: the settled tool row renders the editor's own view, a human interacts with it directly, and anything app-only (the demo's increment_counter is the small example) runs through the Host rather than through the model, so the model never gains a tool it should not have. Because the Host owns the MCP connections, adding an editor is a Host-side configuration rather than a change to the agent loop. The honest scope statement is in the repository: a working editor still needs its own MCP server, which is why the demos are a counter and a separate Three.js editor package.

Tools & Capabilities ★ 1
View on GitHub ↗

Install

dsh plugin --profile web add @creative-dswork/dsh-uni-editor

The README's Install line for the Web profile. The documented way to try it without publishing anything is the bundled standalone counter demo, run against the published Harness CLI rather than a neighbouring source checkout: pnpm run build, then pnpm dlx @deepseek-ai/dsh@0.1.0-rc.6 plugin --profile web add "$PWD", then pnpm dlx @deepseek-ai/dsh@0.1.0-rc.6 web --patch "$PWD/demo/cordis.patch.yml"; opening the printed URL and asking the model to show the counter settles a tool row at 0, and its + button calls an app-only tool through the Host to update the View to 1. The full editor example needs threejs-editor-mcp installed and configured separately.

Compatibility

The runtime is a Cordis plugin bundle: the Host owns its MCP connections, exposes model-visible tools through the harness and keeps app-only tools out of the model registry, so no agent-loop change and no external MCP proxy are required. Untrusted Views are served through a different-origin Sandbox Proxy rather than from the host origin, which is the security boundary the README documents in its own section. One setting needs a decision rather than a default: forwardWorkspace is disabled by default and should only be enabled for a trusted local editor, because it hands the editor the session's workspace. The demo is pinned to Harness 0.1.0-rc.6 in its published-CLI commands, so treat the pin as the tested combination rather than a floor.

Details

Alternatives

dshworks/dsh-solidworks · NaivG/dsh-network · AbnerAI/dsh-monitor

More plugins in Tools & Capabilities

Browse more in Tools & Capabilities

Guides for Tools & Capabilities plugins