DeepSeek Harness Plugin Hub

Publish and manage complete Harness Profiles. Discover Plugins for your next setup.

Explore

PluginsPresetsDocsNews

Community

Publish a pluginContactReport an issue

Resources

Plugin Hub on GitHubDeepSeek HarnessSystem statusPrivacy notice
© 2026 DeepSeek Harness Plugin HubPowered byPaxTech

Independent and unofficial. Not affiliated with, authorized by, or endorsed by DeepSeek.

Cited — DSH Plugin for DeepSeek Harness
← Plugins
C

dsh-cited

Cited

Search the documents of a Cited installation and answer from them, with citations.

The plugin will be installed here. Keep web if you are unsure.

npx -y @deepseek-ai/dsh plugin --profile web add github:RonnieGex/dsh-cited#5df1925c2a3762c7623d0b04d25a85bdf4e5be02
READMECompatibilityVersions

Compatibility and provenance

Cited is published as dsh-cited and currently resolves to version 0.1.0. The Hub verifies its manifest and preserves the exact installation source for reproducible installs.

DSH compatibility
*
Runtime surfaces
any
Release source
github
Registry updated
10/9/2026

Versions

0.1.0stable
10/9/2026

Related plugins

Loading related plugins…

Latest
0.1.0
DSH
*
HMR
Process restart
Tree shaking
Safe tree shaking not declared
Unpacked size
Unavailable
Files
Unavailable
Surface
any
License
Apache-2.0
Source
github
GitHub
★ 0
Weekly downloads
0
Last push
10/10/2026
View source ↗
README badge

Click the badge to copy Markdown for your README.

Do you maintain this Plugin?Claim benefit · Priority security scan

Verify the GitHub repository declared in package.json to manage this listing. After you claim it, Hub will prioritize a security scan of the current version and publish the result when it passes.

Claim this Plugin →
Report an issue
DeepSeek Harness Plugin Hub
ProfilesPluginsCategoriesNewsDocsSign inManage Profiles
ProfilesPluginsCategoriesNewsDocsSign in

Related plugins

More verified plugins in memory-context.

Hindsight Coding Agents@vectorize-io/hindsight-coding-agentsReflect-only Hindsight long-term memory for coding agents (harness-pluggable: opencode, …), with automatic background ingestion (no setup CLI).Memsearch Dsh@zilliz/memsearch-dshMemSearch plugin for DeepSeek Harness: shared markdown memory across agents, with capture, pre-step context injection, memory-recall skill, and a skill-candidate review panel.Memory@furongjun1999/dsh-memoryLingshu (Lingshu·líng shū) DeepSeek Harness plugin: a complete brain—long-term memory/knowledge flywheel/self-awareness/recursive reflection integrated with DSH, with conversations automatically distilled into the md_cg cognitive graph (md documents)Archive Manager@michengai/dsh-archive-manager归档、恢复、收藏和永久删除会话,并诊断损坏的会话日志。 · Archive, restore, favorite, and permanently delete sessions, and diagnose damaged session logs.

README

dsh-cited

Cited inside DeepSeek Harness. Search the documents of a Cited installation and answer from them, with numbered citations, from any agent of the harness.

Cited already speaks the Model Context Protocol over POST <url>/api/mcp. This plugin is the native alternative: one step to install, one card to configure, and two tools (cited_search and cited_ask) that behave like any other DSH tool — no MCP client row in cordis.yml.

Install in one step

In DeepSeek Harness, open Plugins → Add plugin and paste either of these:

github:RonnieGex/dsh-cited
https://github.com/RonnieGex/dsh-cited

The bundle ships its compiled lib/ in the repository, so the installation compiles nothing and asks for no allowBuilds permission. The card Cited appears with the description of the package.

Configure it

Open the card's configuration and fill in two fields:

FieldWhat it is
urlThe address of your Cited installation, like https://cited.example.com. The plugin calls <url>/api/mcp.
tokenThe CITED_MCP_TOKEN of that installation. The field is marked as a secret, so the harness never sends it back to the browser.
timeoutMsHow long one call may take. 30 seconds by default.

An installation without a token has its MCP endpoint off. Start Cited with a long random value and give that same value to this plugin:

openssl rand -base64 32
CITED_MCP_TOKEN=<that value> npm start

Both fields start empty on purpose: an invalid configuration would fail the load of the plugin, and you install before you configure. Until you fill them in, the two tools answer an error that names the missing field and nothing else breaks.

The two tools

cited_search

{ query: string, limit?: 1..8 } → the passages of the documents that match, best first, each with its document, its section, its position and its text. It never writes an answer and never calls a language model, so a search spends none of your balance. The model receives the passages numbered and cites them as [1].

Search the price list of the business.

cited_ask

{ question: string, sessionId?: string } → answered with the answer and its numbered citations, or refused with the honest sentence when the documents do not hold the answer. With a sessionId the turn is stored like any other conversation of the installation.

Ask Cited how much a bicycle tune-up costs and where you got that from.

What the plugin does with your token

  • The token is sent only in the Authorization: Bearer header of a request to the url you configured. The plugin opens no other connection.
  • The token is never written to a log, to a tool result, or to the transcript.
  • A 401, a 403, a 404, a timeout and an unreachable host become short sentences the model can read. None of them contains the token.
  • Failures of Cited arrive as a tool error, never as a crash of the harness.

Development

The plugin is plain ESM JavaScript. src/ is the source and lib/ is the compiled artifact that ships in the repository:

node scripts/link-host-deps.mjs   # links the host packages of your DSH installation
npm run build                     # src/ -> lib/
npm test                          # unit tests and a real Loader composition
npm run gate                      # every claim of the contract, ending in GATE: GREEN

link-host-deps.mjs uses the DeepSeek Harness installation at $DSH_INSTALL, at ../deepseek-harness, or at ../../deepseek-harness. The tests import the host copy of @deepseek-ai/dsh-tools and @deepseek-ai/schemastery, so they exercise the same revision your harness runs.

The gate also installs the plugin from its local path into a temporary DSH_HOME (never your own), boots the headless profile there, reads the plugin card back from the plugin manager, and calls cited_search against a Cited installation served from a sibling worktree on port 3231. Set CITED_REPO to that worktree and DSH_INSTALL to your harness when the two do not sit where the defaults look for them.

What was tested

Verified on Windows on 2026-10-09 against a local Cited with its sample documents:

DeepSeek HarnessResult
Desktop app 0.2.0-rc.2 (its bundled dsh command)dsh plugin add github:RonnieGex/dsh-cited installs in 4 seconds with no build step and no version exemption, the profile composes the dsh-cited layer, and a run with the DeepSeek model called cited_search and answered "a bike tune-up costs 380 pesos [1]" citing cafe-la-horquilla.md
0.1.6-alpha.2 built from sourceinstalls, composes and lists the card; cited_search returns the passages

The unit tests and a composition through the real Cordis loader run on every change. Not verified yet: the same install clicked through the desktop app's Plugins page (it runs the same plugin manager as the command), and cited_ask through the plugin with a chat model.

License

Apache-2.0. See LICENSE and NOTICE. Built by Katalis.

Brief versions of this document: Español · 中文.