wanikua

wanikua / tiangong

Public

天工开物 — 朝廷框架Harness Engineering。三省六部制CLI版。

danghuangshang.com ai aiagent aiagentsframework harness-engineering
15
3
100% credibility
Found Apr 04, 2026 at 15 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
JavaScript
AI Summary

TianGong is an interactive AI framework that simulates ancient Chinese government structures with self-improving agents for collaborative coding and problem-solving tasks.

How It Works

1
🔍 Discover TianGong

You stumble upon this delightful AI helper on GitHub that turns coding into an ancient Chinese court adventure.

2
📥 Bring it home

With a few easy steps, you welcome the AI court onto your computer, ready to serve.

3
👑 Assemble your court

Choose your favorite AI advisor, and watch as wise officials line up in your screen, eager to obey.

4
💬 Give your first command

Simply chat like to a friend—say 'make me a login page'—and the ministers spring into action.

5
⚔️ Enjoy the show

See officials debate ideas, compete for the best plan, or team up to perfect your project.

🏆 Rule with perfect results

Your app or code comes alive, flawlessly built, reviewed, and ready to conquer—just like an emperor's dream.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 15 to 15 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 tiangong?

Tiangong is a JavaScript CLI framework for running self-evolving multi-agent AI systems, modeled after ancient Chinese imperial courts like Tiangong Kaiwu with planning, review, and execution hierarchies. Developers get an interactive REPL where agents handle tasks via natural language prompts, supporting Ollama for local runs or cloud LLMs like Claude. Key commands trigger agent PK battles (/pk), debates (/debate), collaborative coding (/collab), and dream predictions (/dream) on git state or TODOs.

Why is it gaining traction?

It stands out from flat tools like Aider or Cursor by organizing agents into swappable regimes (Ming for speed, Tang for checks), with self-optimization loops that rewrite prompts based on past failures. Gamified hooks like MBTI personalities, reputation ranks, and a treasure hunt drive repeat use, while Viking-style tiered memory slashes token costs by 80%. Natural language routes commands semantically, no memorization needed.

Who should use this?

Solo full-stack devs prototyping apps needing multi-role AI (e.g., code + security + tests in one go). AI tinkerers experimenting with agent societies or competitions. JS CLI enthusiasts bored of single-model tools, especially those running local models on laptops.

Verdict

Fun, opinionated take on multi-agent workflows—try the REPL for agent PKs if you're into JS AI experiments. But with 15 stars and 1.0% credibility, it's alpha-stage: thin docs, no tests visible, risky for real projects. Fork and contribute if the court vibe clicks.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.