Offline mirror of the official DeepSeek Harness documentation as a dsh skill: 168 site pages (EN/zh) plus 52 repository-only reference docs, with topic navigation and a full-text search index for plugin development.
Install
# from npm (prebuilt)
dsh plugin --profile web add dsh-plugin-dev-kb
# from GitHub (first run asks for allowBuilds approval — follow the hint, retry)
dsh plugin --profile web add github:Pasumao/dsh-plugin-dev-kb
Any plugin you install runs third-party code with your own permissions — it can read your files, use your credentials, and reach the network, and tool approvals don’t sandbox it. GitHub-sourced plugins also run build scripts at install time — pnpm blocks those until you allow them, so an install can stop with ERR_PNPM_GIT_DEP_PREPARE_NOT_ALLOWED or ERR_PNPM_IGNORED_BUILDS; dsh prints the exact key to add under allowBuilds in your profile’s pnpm-workspace.yaml, and the install works on the next run. Allowing a build is a trust decision: only install sources you trust, and pin a commit (github:owner/repo#sha).
README
dsh-plugin-dev-kb
中文 | English
194-page official mirror · 608 exclusive supplementary docs · weekly auto-sync from upstream · 0 runtime dependencies
The official docs at your side whenever you write dsh plugins: the entire content of the DeepSeek Harness official documentation site
https://deepseek-harness.github.io/deepseek-harness/ is organized into a form natively usable by dsh —
once this plugin is installed, the agent automatically loads the knowledge base when writing plugins and locates docs by task scenario, so you never have to dig through web pages again.
It also ships the dsh_kb_search tool (since 1.6.0): local scored search over the knowledge base that returns matched files, line numbers, and context snippets in a single call,
and the dsh-upgrade-assistant skill (since 1.7.0): when you upgrade dsh, the agent reads the official per-version upgrade guides and persistence-change records, produces a migration checklist, flags plugin compatibility-gate risks, and plans a rollback path.
Since 1.8.0 it also ships a version-aware banner plus the dsh_kb_read tool (tool output flags "knowledge-base snapshot vs host dsh version" drift and points at the matching upgrade guide; line-numbered windowed reading prevents whole-file reads of huge docs) and the dsh-plugin-starter skill (a scaffold guide for creating compliant dsh plugins from scratch: file checklist, marketplace/awesome gates, verification loop).
What is this
- Complete site mirror: the official docs (VitePress) are generated as a projection of the raw Markdown in
docs/of thedeepseek-ai/deepseek-harnessrepository (links rewritten to site routes, verbatim identical to the live site). 194 pages in both Chinese and English. - Supplementary docs: 608 development references from the repo's
docs/that were not published to the site (glossary, defensive patterns, module graph, testing strategy, incident postmortems, i18n specs, upgrade guides, persistence changes, etc.) — this is exclusive content from the official repo that you won't find on the live site. - Agent-friendly: the
skills/dsh-plugin-dev-kb.mdskill lets dsh auto-load it during plugin development tasks, providing the knowledge base location, topic navigation, and search strategy;kb/meta/topics.mdmaps task scenarios to the files to read;kb/meta/search-index.jsonprovides full search coverage. - Local search tool:
dsh_kb_searchperforms scored search over the knowledge base (title / path / body multi-term weighting) and returns matched file paths, line numbers, and context snippets in a single call; supportsscope: api(search only API-signature/catalog docs),lang: en, and an empty query returning topic navigation. The index is built lazily in memory (~60ms / 400 files for the Chinese index), read-only. - Version awareness (since 1.8.0): tool output carries a banner — host dsh version vs the knowledge-base
master snapshot may drift on APIs, with the matching
extra/upgrade-guide/<vX.Y.Z>guide path; since 1.8.3 the desktop form reads the real core version from the shell'sapp.asarfirst (the shared node_modules copy can lag behind the actual desktop core); detection degrades gracefully (no banner, no error). - Windowed reading tool (since 1.8.0):
dsh_kb_readreads originals bypath/offset/limitline windows, with numbered output, whole-file-read protection, path-escape rejection, automatic zh↔en counterpart hints, and continuation pointers.
Features
- Complete mirror of the official docs: the entire content of the deepseek-ai/deepseek-harness official documentation site, organized into a form natively usable by dsh, 194 pages in both Chinese and English, links rewritten to site routes, verbatim identical to the live site;
- Repo supplementary docs: 608 development references not published to the site (glossary, defensive patterns, module graph, testing strategy, incident postmortems, i18n specs, upgrade guides, persistence changes, etc.);
- Weekly auto-sync: a GitHub Action refreshes the knowledge base from upstream master on schedule (see the "kb sync" badge above for the sync date and commit);
- Agent-friendly: the
dsh-plugin-dev-kbskill lets dsh auto-load the knowledge base during plugin development tasks; thedsh_kb_searchtool locates docs in a single call;kb/meta/topics.mdmaps task scenarios to the files to read, andkb/meta/search-index.jsonprovides full search coverage; - Human-usable: browse the
kb/directory directly, or openkb/INDEX.mdto look things up by URL mapping; - Zero runtime dependencies; two tools (
dsh_kb_search/dsh_kb_read), pure local Node implementations (no network, no file writes).
Verify in 30 seconds
After installation, ask the agent in any session:
What changed in the dsh 0.2.0 settings subsystem (SettingsForms) compared to before?
- If the agent uses
dsh_kb_searchand answers with details likeSettingsForms/.volatile()/ revision conflict detection, it is installed correctly; - If it cannot answer or the search tool is missing, the mount did not take effect — recheck the "Install" step.
Configuration
No configuration required — it works right after installation:
- Reads no environment variables, needs no API key / token, writes no config files;
- After mounting, the
skills/dsh-plugin-dev-kb.mdskill enters the available skills list at session startup and takes effect automatically; - Knowledge base updates go through
npm run rebuild-index(see "Updating the knowledge base"); there are no manual configuration items.
Install / Mount
# npm (recommended)
dsh plugin --profile web add dsh-plugin-dev-kb
# or GitHub
dsh plugin --profile web add github:Pasumao/dsh-plugin-dev-kb
Install from source (local development / debugging):
git clone https://github.com/Pasumao/dsh-plugin-dev-kb.git
cd dsh-plugin-dev-kb
npm install
# mount into the profile as a link: dependency
This plugin is a "pure data + skill" plugin and does not ship with automatic bundle mounting. After installation, add one line at the end of the profile's
cordis.patch.yml to mount it manually (the only one-time step, just copy it;
it mounts as a plugin row in the profile root layer, and its skills/ skill enters the global layer):
- insert:
- id: dsh-plugin-dev-kb
name: dsh-plugin-dev-kb
Only after restarting / creating a new dsh session will the
dsh-plugin-dev-kbskill appear in the available skills list (the skill catalog is snapshotted at session startup).
5-minute verification
After restarting, open a new session and simply ask:
Help me write a minimal dsh plugin
The agent will automatically load the dsh-plugin-dev-kb skill (visible in the skill catalog), locate the docs via the dsh_kb_search tool
(or kb/meta/topics.md), and only then start working — if you see it referencing kb/site/... paths, the verification has passed.
Directory structure
dsh-plugin-dev-kb/
├── cordis.patch.yml mount declaration
├── package.json plugin metadata
├── skills/
│ └── dsh-plugin-dev-kb.md ★ knowledge base usage guide (the skill the agent loads)
├── kb/
│ ├── site/ site mirror: guide/ develop/ reference/ (+ en/ English site)
│ ├── extra/ repo supplementary docs: glossary, defensive-patterns, module-graph, postmortem/, i18n/ …
│ ├── meta/
│ │ ├── topics.md ★ topic navigation: task scenario → files
│ │ ├── search-index.json full index (805 files)
│ │ ├── source.json source commit / time / stats
│ │ └── site-pages.txt list of live site pages
│ ├── INDEX.md site URL ↔ local file mapping
│ └── README.md knowledge base overview and update instructions
└── LICENSE
Usage
- Agent side: when writing plugins / Tools / config / services / events / packaging / LLM adapters, load
the
dsh-plugin-dev-kbskill → prefer thedsh_kb_searchtool (one call returns matched files + line numbers + snippets;scope: apiforctx.*API signatures) → read the source locally by line number; fall back to greppingkb/if the tool is unavailable. - Human side: browse the
kb/directory directly, or openkb/INDEX.mdto look things up by URL mapping.
Self-check
Run the structural self-check before publishing (offline, zero dependencies):
npm run selfcheck # structural integrity + entry / bundle patch existence
Updating the knowledge base
See kb/README.md: re-clone deepseek-ai/deepseek-harness (master branch), install the projection dependencies, then
call projectDocs() from scripts/project-doc-site.ts inside the repo to generate website/.generated/,
overwrite this plugin's kb/site/ (and sync kb/extra/ according to the publishing checklist), then run
node scripts/rebuild-index.mjs in this plugin's root directory to rebuild the index and INDEX.md.
Related plugins
This plugin is part of Pasumao's dsh plugin ecosystem; other published plugins in the same series can be used alongside it:
| Plugin (npm) | GitHub | Description |
|---|---|---|
| dsh-notify | GitHub repo | Native Windows notifications + system tray |
| dsh-plugin-choice-refresh | GitHub repo | Choice enhancement: regenerate options / more options |
| dsh-plugin-image-tools | GitHub repo | Image choice cards + inline images in replies + image handoff for blind models |
| dsh-plugin-table-zoom | GitHub repo | Floating viewer for long chat tables + one-click copy as Markdown |
| dsh-plugin-windows-guard | GitHub repo | Windows environment pitfall prevention: rule skills + mojibake detection / dangerous-write interception / encoding diagnosis & repair |
| dsh-plugin-workbench | GitHub repo | VS Code-style file explorer + editable preview |
| dsh-plugin-context-trim | GitHub | Per-session injection gate: trim skills / tools / prompt sections |
See Pasumao · dsh plugins for the rest of the series; if you find them useful, a ⭐ on GitHub is much appreciated.
AI-generated content statement
The knowledge base content is an organized mirror of the official documentation (sources noted in kb/meta/source.json); the index and
navigation were AI-assisted (DeepSeek Harness), and all of it has been manually verified.
License
MIT
Links
More in this category
tt-a1i/archify#integrations/deepseek-harness★ 98372
Generate validated, self-contained interactive architecture, workflow, sequence, data-flow, and lifecycle diagrams from repositories or system descriptions.
dream-num/dsh-univer-office★ 742
Give DeepSeek Harness a real office environment. Univer Office Plugin brings spreadsheets, docs, slides, canvases, relational tables, and more into one runtime — with connected data, validation, versioned changes, and isolated worktrees for multi-agent collaboration.
Roland0511/dsh-marktext★ 388
WYSIWYG Markdown editor in the DSH right sidebar, built on MarkText's muya engine: open a .md file, edit it as a rendered document, and save back through ctx.fs with a version guard.
ECNU/EduWork#dsh-knowledge-studio★ 383
A workspace-based Studio for source-grounded reports, slides, tables, mind maps, quizzes, flashcards, audio and video, using DSH models and shared artifact services. Office generation requires Python 3.12+ with its dependencies and DSH_OFFICE_PYTHON configured; media rendering requires Chromium, and local ASR additionally requires whisper.cpp and a model.
PerryLink/dsh-industry-research★ 225
Deterministic industry research reports for DeepSeek Harness — company and industry research flows produce structured, verifiable reports from staged evidence.
joeseesun/qiaomu-ui-learn-dsh★ 87
Browse UI design references and interaction examples, take design quizzes, and export learning notes as Markdown.
Community comments
Comments are public GitHub Discussions. Loading them connects to GitHub and Giscus; a GitHub account is required to post.