Biajin-PKU

Biajin-PKU / research-harness

Public

面向科研文献工作的 Agent Harness：持久化 SQLite 状态、69 个类型化原语、112 个 MCP 工具、6 个证据门禁阶段，每次经记录的调用都有溯源。可由 Claude Code / Codex / Python / rh CLI 驱动。

10
0
100% credibility
Found Apr 23, 2026 at 10 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

Research Harness is a guided AI workflow that automates literature review, gap finding, proposal generation, experiment design, and paper drafting while keeping everything traceable and reviewable.

How It Works

1
🔍 Pick your research idea

Tell it about your topic, like 'diffusion models for bidding', and add a few starting papers you know.

2
🚀 Launch the research engine

Hit start and watch it automatically find related papers, read them deeply, and build a knowledge map.

3
📊 Review the insights

See summaries of key claims, gaps in the field, and promising research directions it discovered.

4
Choose a direction
✅
Approve and continue

Green light the best direction to build experiments.

✏️
Edit or restart

Adjust the plan and rerun parts that need work.

5
🧪 Run experiments and draft

It designs tests, runs them safely, and writes your paper sections with full traceability.

🎉 Get your polished paper

Review the complete draft with checks for accuracy, then export ready for submission.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 10 to 10 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is research-harness?

Research-harness is a Python agent harness for scientific literature workflows, turning LLMs into reliable research assistants for lit reviews, proposals, experiments, and drafting. It persists everything in a single SQLite database—papers, claims, gaps, artifacts—with full provenance on calls, and gates progress through six stages (init to write) based on typed evidence. Drive it via rh CLI, Python API, MCP tools for Claude Code or Codex, or a web dashboard.

Why is it gaining traction?

Unlike generic agent frameworks, it enforces auditability and continuity across weeks-long projects, resuming seamlessly across Claude, OpenAI, or CLI tools like agent GitHub Copilot. The 112 MCP tools from 69 primitives make it plug-and-play for agent GitHub Claude setups, with skills for lit mapping, gap detection, and drafting. Early buzz on agent harness Anthropic patterns and GitHub agent repo benchmarks draws devs building domain-specific agents.

Who should use this?

PhD students or applied ML researchers drowning in lit reviews needing structured claims and gaps. Agent engineers prototyping research harnesses on GitHub Copilot CLI or VSCode. Reproducibility teams in labs wanting traceable numbers from experiments to drafts, like dog harness research for educational pipelines.

Verdict

Solid 0.1.0 with 987+ tests and bilingual docs, but 10 stars and 1.0% credibility signal early days—fine for prototypes, risky for production. Grab it if lit-heavy agent workflows fit; fork and contribute to mature the GitHub agent HQ reference. (198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.