dsh-web-search-ddg
aooyoo
Zero-token DuckDuckGo search provider for the DeepSeek Harness (DSH) web seam — local headless browser, no API key, no m…
PROJECT TOPICS
PROJECT README
English | 中文
A DeepSeek Harness bundle plugin that adds the MCP Servers and Skills settings pages to the dsh web UI.
DeepSeek's npm-shipped dsh (npm install -g @deepseek-ai/dsh) ships an older
web-app bundle that omits these two settings pages. This plugin backports
them into your web profile with one command — you keep every other part of
your dsh install untouched.
SKILL.md. The
host writes uploaded skills to $DSH_HOME/skills/<name>/SKILL.md, where the
built-in filesystem skill provider auto-discovers them.The underlying MCP client, skill loader, settings seam, and connection status are all provided by dsh itself — this plugin contributes only the settings UI pages. It does not re-implement or replace any of dsh's MCP or skill mechanisms.
Install into your web profile. This plugin is published to npm, so the
simplest path is the package name (no github: needed):
dsh plugin --profile web add dsh-settings-mcp-skills
On install pnpm runs a small postinstall script that ships with the bundle.
It does no compilation and reaches no network — it only extracts the two
pre-built official packages (bundled as tarballs under deps/) into your
profile's node_modules so the settings pages can load.
pnpm (≥10) blocks build scripts (postinstall/prepare) until they are
allowlisted — for both npm and git installs. On the first run, pnpm prints
the exact package key and the file to edit. Set that key to true under
allowBuilds in your profile's $DSH_HOME/profiles/web/pnpm-workspace.yaml,
then re-run the same add command. You only do this once.
pnpm 11 release-age gate: for ~10 days after a new version ships, pnpm may
silently install an older version instead (e.g. add resolves 0.1.0 while
latest is 0.1.1). If dsh plugin ... add reports +1 but you do not see
postinstall run (or the settings pages still do not appear), add the version
to your profile's pnpm-workspace.yaml and re-add:
minimumReleaseAgeExclude:
- dsh-settings-mcp-skills@0.1.1
Then dsh plugin --profile web add dsh-settings-mcp-skills@0.1.1.
Once installed, boot and the two new pages appear under Settings:
dsh --profile web
To uninstall:
dsh plugin --profile web remove dsh-settings-mcp-skills
Pin a version for repeatability: dsh-settings-mcp-skills@0.1.0.
You can also install from the GitHub repository or a local clone/tarball:
dsh plugin --profile web add github:w123456789zy/dsh-settings-mcp-skills
dsh plugin --profile web add ./dsh-settings-mcp-skills
# or
dsh plugin --profile web add ./dsh-settings-mcp-skills-0.1.0.tgz
The allowBuilds step applies to these too (pnpm gates postinstall
regardless of source). Pin a git commit for repeatability:
github:w123456789zy/dsh-settings-mcp-skills#<sha>.
The bundle inserts two plugin rows into the web profile and ships the two packages they reference, so they resolve locally without needing npm:
| Row id | Package | Role |
|---|---|---|
ui-settings-tools |
@deepseek-ai/dsh-client-ui-settings-tools |
Browser settings pages (MCP + Skills), registered as settings.section slots |
host-tool-settings |
@deepseek-ai/dsh-host-tool-settings |
Host service owning mcp-servers / skill-sources namespaces + disk projection |
The cordis.patch.yml layer is applied after web-app, so it only adds rows
that the npm-shipped web-app leaves out and never overrides your own profile
or home-level cordis.patch.yml.
This bundle itself is MIT. It re-distributes two DeepSeek Harness packages
(@deepseek-ai/dsh-client-ui-settings-tools and @deepseek-ai/dsh-host-tool-settings),
which are DeepSeek components licensed MIT, Copyright (c) 2026 DeepSeek.
They ship as bundled tarballs under deps/, and each tarball carries its own
unchanged MIT LICENSE. THIRD_PARTY_NOTICES.md
records the attribution. This re-distribution is permitted by their MIT license.
No build step is needed for users — the dependencies ship pre-built. To package your own tarball before publishing:
pnpm pack CLASSIFICATION EVIDENCE
系统优先读取 GitHub Topics,再与站内分类词典和词根规则比对。