Opt-in web consultation for DeepSeek Harness tasks: paste the material and a first-pass review, let a sub-agent check one specific doubt against a web chat, and keep the verification with the agent. Its HTTP routes have no DSH access check, so a web page open in your browser could call them.
Install
# from npm (prebuilt)
dsh plugin --profile web add @dsh-llsa/web-collab
# from GitHub (first run asks for allowBuilds approval — follow the hint, retry)
dsh plugin --profile web add github:TaoruiLiu19/DSH-LLSA#path:/packages/web-collab
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
A second opinion before you commit — without giving a web chat your files.
An opt-in plugin for DeepSeek Harness: it opens a small panel where you paste task material and a first-pass review, and the agent may consult a web chat to check a specific doubt. The agent keeps the job of verifying the source material and the deliverable — a reply is an opinion to check, not a result to trust.
It is off until you turn it on, it only runs behind a loopback web server, and nothing is sent anywhere unless you paste it.
What actually happens
| Shape | What it does |
|---|---|
| Single agent | Paste text or a UTF-8 attachment, add your first-pass advice, and the Harness makes one generation request to check it and deliver |
| Multi-agent | Paste the task material; a sub-agent may decide on its own to check a doubt. At most one consultation per attempt, question capped at 1024 UTF-8 bytes. The short reply rides back with the tool result; the full Markdown stays in the local collaboration record |
| Public chat page | Manual copy and paste. A sub-agent's question keeps its consultation id on the first line of the Markdown. With Open WebUI configured, replies are fetched from its official API |
Turning it on
The bundle ships it disabled. Start DSH-LLSA with the flag:
DSH_LLSA_WEB_COLLAB=1
Open the terminal's Harness auth link, then go to the same address at /web-collab. Pick a task, tick 网页协作, and only then does a consultation happen. Clear the variable (or set it to 0) and restart to turn the built-in copy off again.
To fetch replies from Open WebUI automatically, set these before starting:
DSH_WEB_COLLAB_OPEN_WEBUI_URL=…
DSH_WEB_COLLAB_OPEN_WEBUI_MODEL=…
DSH_WEB_COLLAB_OPEN_WEBUI_KEY=…
A non-loopback instance must use HTTPS.
What leaves the machine
- Only what you paste or attach. The panel does not read your repository, and the plugin does not sample your session.
- A loopback web server is required.
apply()refuses to mount on a non-loopback host. - Requests are JSON, capped, and cached by nobody.
Content-Type: application/jsonis enforced, bodies are limited to 128 KiB, responses carryno-storeand a strict CSP, and the panel is framed byframe-ancestors 'none'. - Cost sits with the consulted provider. An Open WebUI model bills its own service; whether the whole job gets cheaper depends on whether the advice replaces other model calls. This plugin does not claim a saving.
Install into Harness
Requires Node.js 24.11+, DeepSeek Harness 0.2.2-rc.1, and pnpm for dsh plugin.
From the repository root:
npm pack ./packages/web-collab
dsh plugin --profile web add ./dsh-llsa-web-collab-0.2.0.tgz
dsh web --host 127.0.0.1 --port 3095 --no-open
Once published, the package name works too:
dsh plugin --profile web add @dsh-llsa/web-collab
A standalone install is managed by Harness's web profile. If you also use the copy bundled with DSH-LLSA, remove one of them first — two copies load twice.
Development
npm --prefix ./packages/web-collab install # pulls the two peer dependencies
npm --prefix ./packages/web-collab test # 19 tests
Peer dependencies are @deepseek-ai/cordis@4.1.0 and @deepseek-ai/dsh-llm@0.2.2-rc.1; both are provided by a Harness install, so the package ships with no runtime dependencies of its own.
| Path | Responsibility |
|---|---|
src/index.js |
The service, the /web-collab routes, body limits and headers |
src/panel.* |
The panel UI: tasks, advice, consultation list |
src/single-agent.js |
Single-agent job lifecycle and its one generation request |
src/child-consultations.js |
Per-attempt child consultations and their caps |
src/browser-advisor.js |
Manual exchange and the Open WebUI client |
src/native-handoff.js |
Handing evidence to native tickets |
src/task-evidence.js |
Evidence validation and size limits |
License
MIT.
Links
More in this category
loopx-project/loopx#dsh-loopx-plugin★ 6694
LoopX, a provider-neutral, local-first state kernel and control plane for long-horizon agents: keeps Goal, Todo, gate, evidence, quota, recovery, and handoff state above DeepSeek Harness, while the plugin bootstraps the CLI and skills, admits bounded same-session continuation, and adds a loopback GoalBar for the exact bound loop.
Q00/ouroboros#integrations/dsh-plugin★ 6373
Config-only bundle that mounts Ouroboros through the DSH MCP client, exposing 36 interview, Seed, execution, evaluation, and evolution workflow tools in DSH.
chuspeeism/dashi-taskboard#deepseek-harness★ 3495
Embeds the active installed Codex Taskboard runtime in the DeepSeek Harness sidebar, using its launcher runtime descriptor instead of a fixed port.
NanmiCoder/dsh-agent-teams★ 2978
AgentTeams multi-agent teams.
EthanYoQ/AI-Novel-Writer#dsh-ai-novel-writer★ 2670
Installs a dedicated AI novel-writing preset and workbench: revisioned local project assets, a compact side drawer, and native approval-gated single-file changes.
shanliuling/dsh-image-gen★ 1576
Native conversational image generation for DeepSeek Harness: ask the agent to create an image, and it handles generation and keeps the result directly in the conversation.
Community comments
Comments are public GitHub Discussions. Loading them connects to GitHub and Giscus; a GitHub account is required to post.