MIT-licensed Python multi-agent literature-review framework that runs as a native DeepSeek Harness plugin — requirement interview, 10+ adversarial debate departments, per-claim confidence scores, and NLI citation verification, with a DSH control panel for full-text upload and re-verify.
DSH integration
Native runtime
Author-claimed
Safety audit
Unaudited
Last verified
2026-09-02
License
MIT
01What can it help you accomplish?
Drive a multi-agent literature review for a research topic directly from a DeepSeek Harness chat
A structured literature review with per-claim confidence scores, methodology comparison matrices, and NLI-verified citations — exported as Markdown + DOCX
Researchers, graduate students and academic teams who need a rigorous review and want to fight single-model hallucination
Supervise the debate in real time and batch-upload paywalled full texts for NLI re-check from the DSH console
A citation-verification report upgraded with full-text NLI (verified / mismatch / overstated) and your imported papers force-included in the final report (PDF / DOCX)
Users running the native DSH plugin who need to close the paywalled-paper evidence gap and re-verify claims
02How to install into DeepSeek Harness
Prerequisites
- Python 3.10+ (3.11+ recommended) — the README lists it as a requirement
- A DeepSeek API key (about $0.05–0.10 per full run; any OpenAI-compatible endpoint also works)
- git — required for the one-shot installer / clone
- Internet access to the DeepSeek API (custom endpoints supported); no GPU and no database needed
Installation steps
- 01
Clone the project: `git clone --depth 1 https://github.com/fangqian616/consensus-pipeline.git`
$ git clone --depth 1 https://github.com/fangqian616/consensus-pipeline.git
- 02
Install the DSH plugin bundle (auto-registers native tools + the /consensus-pipeline/ panel): `npx -p @deepseek-ai/dsh dsh plugin --profile web add file:./consensus-pipeline/dsh-plugin`
$ npx -p @deepseek-ai/dsh dsh plugin --profile web add file:./consensus-pipeline/dsh-plugin
- 03
Set your DeepSeek API key (required): `export DEEPSEEK_API_KEY="sk-your-key-here"`
- 04
In chat, tell the agent your research direction — it runs the requirement interview, then starts the pipeline
- 05
Open the 📊 控制台 floating button (bottom-right) to watch progress, upload full text, and re-verify
- 06
Alternatively, run the one-shot installer for a non-DSH MCP client: `curl -fsSL https://github.com/fangqian616/consensus-pipeline/raw/main/install.sh | bash`
$ curl -fsSL https://github.com/fangqian616/consensus-pipeline/raw/main/install.sh | bash
Verify the integration
- After first use the 📊 控制台 floating button appears bottom-right and the plugin auto-clones the full project to ~/.dsh/consensus-pipeline
- On first chat use the agent runs the requirement interview and then launches the multi-round debate, shown live in the panel
03DSH integration and capability boundaries
Native DSH plugin bundle — `npx -p @deepseek-ai/dsh dsh plugin --profile web add file:./consensus-pipeline/dsh-plugin` auto-registers native tools plus the /consensus-pipeline/ control panel; on first use it auto-clones the full project to ~/.dsh/consensus-pipeline.
Multi-agent debate pipeline
a research topic (or requirement-interview answers)→structured literature review with per-claim confidence scores; debate results structured into JSON with roles, positions and consensus points
Multi-source paper search + 3-layer QC
a research topic; OpenAlex + Semantic Scholar + arXiv→deduplicated, abstract-backfilled papers passing a 3-layer QC sieve (hard filter → LLM classify → importance tagging)
Per-claim confidence annotation
debate consensus points→each conclusion tagged with how many papers support it (e.g., "42/77 papers, high confidence")
DSH control panel
paywalled PDFs / pending full-text imports→live progress, atomic-verification card, full-text batch upload, one-click re-verify with progress bar, pending-import list (pause / continue / skip)
04Who is it for? When not to use it?
Good for
- Researchers, graduate students and academic teams who need a rigorous review and want to fight single-model hallucination
- Users running the native DSH plugin who need to close the paywalled-paper evidence gap and re-verify claims
Not for
- A full run takes 10-30 minutes and about $0.05-0.10 in DeepSeek API costs; more departments mean more API calls. No GPU needed.
05Compatibility, maintenance and safety notes
- A full run takes 10-30 minutes and about $0.05-0.10 in DeepSeek API costs; more departments mean more API calls. No GPU needed.
- Needs Internet access to the DeepSeek API (custom OpenAI-compatible endpoints supported). Paper retrieval uses free open APIs (arXiv / Semantic Scholar / OpenAlex); no database needed.
- Some UI labels are bilingual (Chinese/English mix) even in English mode — the maintainer lists this as still rough.
MIT · active (latest release v0.12.18, 2026-08-01 per repo metadata); curated in the DSH catalog
06Frequently asked questions
How do I install Consensus Pipeline into DeepSeek Harness?
Clone the repo, then run the DSH bundle command `npx -p @deepseek-ai/dsh dsh plugin --profile web add file:./consensus-pipeline/dsh-plugin`. It auto-registers native tools and the /consensus-pipeline/ panel; on first use it auto-clones the full project to ~/.dsh/consensus-pipeline.
Does it run as a native DSH plugin or only over MCP?
Native. The README documents a DSH entry with native tools + control panel, separate from the zero-dependency MCP server for Claude Desktop / Cursor / Codex. The DSH bundle mounts the /consensus-pipeline/ panel and registers native tools directly.
What can I do from the DSH control panel?
The 📊 控制台 panel shows live progress, an atomic-verification card, full-text batch upload (DOI auto-extracted from PDFs), one-click re-verify with a progress bar, and a pending-import list with pause / continue / skip.
How do I add my own paywalled papers inside DSH?
Use the DSH control panel to batch-upload paywalled PDFs (any filename; DOI auto-extracted), then one-click re-verify — imported papers are force-included in the report. You can also drop PDFs into fulltext_papers/ and run --rerun-67.
Is a DeepSeek API key required, and how much does a run cost?
Yes — DEEPSEEK_API_KEY is listed as required. A full run (e.g. 148 papers, 10 departments, 2-3 rounds) costs about $0.05-0.10 and takes 10-30 minutes; any OpenAI-compatible endpoint is supported.
07Related DSH workflows
archify
by tt-a1i
Agent skill for beautiful, verifiable architecture, workflow, sequence, data-flow, and lifecycle diagrams—self-contained HTML with motion and crisp export.
openviking
by volcengine
Self-evolving Context Database for AI Agents. Unify Agent Memory, Knowledge RAG and Skills.
nocobase
by nocobase
NocoBase is an open-source AI + no-code platform for building business systems fast. Instead of generating everything from scratch, AI works on top of production-proven infrastructure and a WYSIWYG no-code interface, so you get both speed and reliability.
learn-harness-engineering
by walkinglabs
Harness engineering beginner tutorial, from 0 to 1
08Data and sources
#### DSH (DeepSeek Harness) — native tools + control panel
Install as a bundle (auto-registers native tools + the `/consensus-pipeline/` panel):
On first use, the plugin auto-clones the full project to `~/.dsh/consensus-pipeline`
This page is generated from the project’s public documentation, repository metadata and a structured parse of DSH Plugins; last verified on 2026-09-02. Found an error? Submit a correction.
Best DeepSeek Harness Plugins
Twelve plugins worth installing first — picked from the whole catalog, across every category.