dsh-enhance

简体中文

A toolkit of native plugins that fill the gaps of the DeepSeek Harness / DeepSeek models: capabilities the official stack does not have or does poorly, as one composition row each, configured entirely from the Web settings UI — no CLI scripts, no manual YAML editing, no npm publishing.

What you get

Capability	Package	Tools	Fills the gap
Vision	packages/dsh-vision	vision_describe, vision_list_models	DeepSeek models have no image input; this gives agents eyes on demand through any Responses API-compatible multimodal model endpoint
Native web	packages/dsh-native-web	native_search, native_scrape	Built-in search depends on a cloud provider (extra token cost); MCP bridges add hops. This talks directly over HTTP to your local self-hosted web instance (Firecrawl-compatible), no bridge, no cloud

Both plugins are host-side Cordis plugins: one fetch per tool call, per-call timeouts, caller-cancellation aware, API keys marked secret and redacted on the wire.

Install

One-time, per machine (the packages are installed locally from this repo — no npm registry account needed):

npm pack ./packages/dsh-vision ./packages/dsh-native-web
npm install -g ./vcxmug-dsh-vision-0.1.0.tgz ./vcxmug-dsh-native-web-0.1.0.tgz

(npm install -g ./packages/... would create symlinks whose dependencies do not resolve; packing first installs real copies with their dependencies.)

The plugins are written in TypeScript (src/index.ts, strict mode); the built lib/ is committed, so packing and installing work with no build step. After editing the source, rebuild with npm install && npm run build inside the package directory.

Then, in the DeepSeek Harness Web UI:

1. Settings → Agent presets — create a preset (or duplicate an existing one) and add the rows you want (copy from presets/vision.cordis.yml and presets/native-web.cordis.yml, or just the two lines):

   - id: vision
     name: '@vcxmug/dsh-vision'
   - id: native-web
     name: '@vcxmug/dsh-native-web'
   

2. Settings → Plugins → dsh-vision / dsh-native-web — configure the form: endpoint base URL, model, API key, instance URL/key, timeouts, …
3. Start a new session with that preset. Changes made later in Settings apply on the next tool call — no restart.

Requirements

• DeepSeek Harness 0.1.0-rc.6+ (the web profile; agent presets under $DSH_HOME/.agent-presets)
• Node 18+ (global fetch)
• Vision: any multimodal model endpoint that speaks the OpenAI Responses API (/responses) — official or third-party endpoints
• Native web: a reachable self-hosted Firecrawl-compatible instance (docker compose, see firecrawl/firecrawl); USE_DB_AUTHENTICATION=false instances accept any API key

Repository layout

packages/dsh-vision/        # vision plugin (TS in src/, built lib/ committed)
packages/dsh-native-web/    # native web plugin (TS in src/, built lib/ committed)
src/dsh-http/               # Go helper binary for the dynamic-plugin form (stdlib only)
presets/                    # pure mount-point composition fragments
docs/                       # known limitations, helper binary, self-test prompt

Notes

• Native web vs MCP: MCP routes each call through the DSH MCP client and a firecrawl-mcp subprocess — extra hops and session turns. The native route is one direct HTTP call to your instance. (Prefer MCP? The upstream @deepseek-ai/dsh-mcp-client works for that.)
• Dynamic-plugin form (session-scoped, no agent preset): the plugin calls the dsh-http Go helper — Go standard library only, zero third-party dependencies — through the shell service with every variable passed via environment/stdin, so the API key never appears in a process argv. See docs/helper-binary.md for the contract and build.
• Known limitations: see docs/known-limitations.md.
• End-to-end verification: docs/self-test-prompt.md.

Testing

npm test runs one runtime verification: a scripted mock LLM (no API key, no network) drives a real headless dsh agent loop with both plugins mounted and asserts the tool loop completes end to end. Requires Node >= 22.18 (native TypeScript execution, no build step) and the dsh launcher on PATH (or the DSH env var); the test skips when dsh is missing.

npm run typecheck checks the test sources; npm run build rebuilds both plugins' lib/ from src/. Both need dev dependencies installed first (npm install at the repo root and inside each package directory).

License: MIT