PaperMate

面向中国大学生的论文写作智能体 · Human-in-the-loop Thesis Copilot

English · 简体中文

---

PaperMate is an early-stage, human-in-the-loop research and thesis-writing copilot for Chinese university students.

It is designed for students who do not yet know how to narrow a topic, search literature, organize evidence, and turn that material into a thesis structure. PaperMate assists the process — it does not replace the student.

Status: Early open-source WIP. The runtime and search layer are usable. End-to-end LLM drafting is in progress.

---

Positioning

PaperMate is not a "write my whole paper" agent.

The first public scope is deliberately narrow:

• Chinese undergraduate and master's thesis workflows
• Human-in-the-loop research and writing assistance
• Evidence-first search, note organization, and draft preparation

The long-term architecture supports other languages and academic workflows, but that is not the current focus.

Who it is for

• Chinese university students who need help learning the paper-writing workflow
• Builders who want an open, inspectable paper-writing runtime
• Researchers exploring controlled, checkpoint-based writing assistance

What it is not

• A fully autonomous ghostwriting tool
• A polished end-to-end thesis product
• A universal solution for every school, field, or citation style

What it does today

• Routes a paper-related goal into a workflow stage
• Runs multi-source literature search across OpenAlex, Crossref, and arXiv
• Fuses, reranks, and deduplicates candidate papers
• Produces structured candidate sets with explicit selection reasons
• Keeps runtime state, checkpoints, and session artifacts organized

Why this project exists

Most writing agents optimize for maximum automation. PaperMate takes a different path:

• Human-in-the-loop — pauses before high-stakes actions
• Evidence-first — claims stay tied to sources
• Workflow-aware — topic framing, search, notes, drafting, review, and revision are distinct stages
• Open and hackable — the runtime, contracts, and prompts are all visible in the repo

This makes PaperMate a better fit for students who need support and structure, but still need to understand and defend what they write.

Current limitations

• The search layer is the most mature part of the system
• End-to-end LLM drafting is not fully wired into the runtime yet
• Chinese thesis templates, school-specific rules, and citation formats are not standardized yet
• Browser-backed retrieval requires local Chrome DevTools setup and explicit opt-in for local bb-sites adapters

Quick start

npm install
npm run test:paper-writer
npm run smoke:paper-writer

Use with Claude Code

For local development, load this repository as a session plugin directory:

claude --plugin-dir ~/PaperMate

This makes Claude Code load:

• /paper-writer-run
• the papermate-* agents
• the paper-writer thesis workflow agent
• the repo-local skills in skills/

If PaperMate is later published to a Claude Code marketplace, install it with:

claude plugin install papermate

Try the runtime in code:

const { runPaperWriterEntry } = require('./scripts/paper-writer-entry');

const result = await runPaperWriterEntry({
  goal: 'review retrieval-augmented generation for biomedical question answering',
  searchMode: 'real',
});

console.log(result.ui.guidance);
console.log(result.runtime.searchArtifact.items.length, 'candidates found');

For browser-backed arXiv retrieval, start Chrome with remote debugging enabled. If you want to execute local bb-sites adapters, explicitly opt in first:

PAPERMATE_ALLOW_UNTRUSTED_BB_SITES=1

Then run:

const result = await runPaperWriterEntry({
  goal: 'review retrieval-augmented generation for biomedical question answering',
  searchMode: 'browser',
  browserUrl: 'http://127.0.0.1:9222',
});

Repository layout

PaperMate/
├── .claude-plugin/         # Claude Code plugin manifest
├── agents/                 # Agent prompts and role definitions
├── commands/               # Slash-command entry points
├── docs/                   # Design docs and runtime contracts
├── scripts/                # Runtime, search, session, and adapter code
├── skills/                 # Auto-activating skill definitions
├── README.md
└── ROADMAP.md

Key files to read first

• agents/paper-writer/paper-writer.md — core workflow prompt
• scripts/paper-writer-search-layer.js — most mature implementation today
• docs/paper-writer/paper-writer-runtime-contracts.md — runtime contracts and data model
• docs/paper-writer/progress-map.md — implementation maturity and gaps

Roadmap

See ROADMAP.md for current priorities, limitations, and next milestones.

Example flow

See examples/cs-master-related-work.md for a complete example covering:

• thesis-oriented routing
• literature search
• candidate set construction
• related-work drafting

Contributing

Issues and PRs are welcome. Please open an issue first to discuss what you would like to change.

License

This project is licensed under the GNU Affero General Public License v3.0 (AGPL-3.0).

This means:

• You can use, study, modify, and distribute this software
• If you deploy a modified version as a network service, you must release your source code under the same license
• Commercial use is permitted, but you cannot make it proprietary

See LICENSE for the full text.