https://github.com/tumblecat44/universe-grok
Truth-seeking orchestration harness for Grok Build. Actively researches current best practices at session start via X and web.
https://github.com/tumblecat44/universe-grok
agent ai-agents grok grok-build harness orchestration subagents truth-seeking xai
Last synced: 11 days ago
JSON representation
Truth-seeking orchestration harness for Grok Build. Actively researches current best practices at session start via X and web.
- Host: GitHub
- URL: https://github.com/tumblecat44/universe-grok
- Owner: tumblecat44
- License: mit
- Created: 2026-05-18T10:53:46.000Z (about 2 months ago)
- Default Branch: main
- Last Pushed: 2026-05-21T05:05:57.000Z (about 1 month ago)
- Last Synced: 2026-05-21T11:53:51.516Z (about 1 month ago)
- Topics: agent, ai-agents, grok, grok-build, harness, orchestration, subagents, truth-seeking, xai
- Language: Shell
- Size: 103 KB
- Stars: 1
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- Contributing: CONTRIBUTING.md
- License: LICENSE
- Governance: GOVERNANCE.md
- Agents: AGENTS.md
Awesome Lists containing this project
README
# Universe Grok
**A truth-seeking orchestration harness for Grok Build.**
Universe Grok turns Grok Build from a powerful model into a **system that actively pursues current truth** instead of relying on stale training data.
It is not just another skill. It is an attempt to build the highest-leverage way to use Grok for serious software engineering work.
## Core Philosophy
> The biggest limitation of current coding agents is not intelligence — it is operating on outdated internal knowledge.
Universe Grok's defining behavior:
- At the start of every session, it proactively researches **current best practices** using X (Twitter) and Web search.
- It treats "what the model already knows" as a hypothesis, not as truth.
- All orchestration, verification, and agent collaboration exists to serve **better contact with reality**.
See [CONSTITUTION.md](./CONSTITUTION.md) for the full principles and Level 2 → Level 3 vision.
## Current Status
- **Target**: Level 2 (Reliable, high-quality harness)
- **Vision**: Level 3 (The standard way serious people use Grok Build)
We are building this in the open as a public project.
## Features (Current)
- Strong curated personas (`universe-verifier`, `universe-researcher`, `universe-tactician`, etc.)
- Session Start automatic best-practice research hook (X + Web)
- **First Questions** companion skill (`/first-questions`) — ruthless "Question every requirement" ritual that forces a one-sentence goal and deletes scope before any work begins
- **First Principles** companion skill (`/first-principles`) — forces explicit sourced bedrock facts and reconstruction from reality (not analogy) before solution work; co-equal highest-authority artifact with first-questions.md
- Clear Constitution that actually influences behavior
- Deep integration philosophy with existing high-quality skills (`/check`, `/best-of-n`, etc.)
## Installation
```bash
# 1. Clone the repo
git clone https://github.com/your-org/universe-grok.git ~/.grok/plugins/universe-grok
# 2. Install the personas (recommended)
cp ~/.grok/plugins/universe-grok/personas/*.toml ~/.grok/personas/
# 3. Restart Grok Build or reload plugins
```
After installation, activate with natural language:
- "universe mode" / "full power"
- "first questions" + "first principles" (the required non-negotiable pair as the very first steps on any serious work)
- "first principles" (the required reconstruction step after first-questions, before orchestration)
- "push Grok to the limit"
Recommended order for maximum-leverage sessions:
1. `/first-questions` — Kill ambiguity and delete everything that fails the test.
2. `/first-principles` (on surviving scope) — Reconstruct the minimal solution from explicit bedrock truths before any design or implementation.
3. research hook if needed
4. `/universe` — Orchestrate with the two sacred artifacts (first-questions.md + first-principles.md) as co-equal contracts.
Or explicitly with `/universe` and `/first-questions`.
## How It Works
**The highest-leverage flow** (use this):
1. **First Questions** (`/first-questions`) — Before anything else, force the One-Sentence Goal and ruthlessly interrogate every requirement using Elon’s "Question every requirement" protocol. Delete what fails. Produce the sacred `.grok/first-questions/first-questions.md` artifact.
2. **Session Start Research** — Universe Grok’s hook triggers fresh X + web research into current best practices.
3. **Orchestration** — Uses strong personas and optimal subagent topologies.
4. **Verification** — Independent `universe-verifier` on every meaningful change.
First Questions is the foundation. Universe Grok orchestration is only as good as the clarity of the goal it receives.
## Project Structure
```
universe-grok/
├── skills/
│ ├── universe-grok/SKILL.md # Main orchestration logic
│ └── first-questions/SKILL.md # Ruthless requirement interrogation (run this first)
├── personas/ # Universe-optimized personas
├── hooks/ # SessionStart truth-seeking hook
├── CONSTITUTION.md # The soul of the project
├── plugin.toml
└── README.md
```
## Roadmap
See [CONSTITUTION.md](./CONSTITUTION.md) for Level definitions.
High priority:
- Stronger Session Start research protocol
- Better automatic stack detection
- More powerful personas
- Richer research artifact consumption by subagents
## Contributing
We welcome contributions, especially in:
- High-signal personas
- Research protocols and truth-seeking mechanisms
- Real usage examples and case studies
Please read [CONTRIBUTING.md](./CONTRIBUTING.md) and [CONSTITUTION.md](./CONSTITUTION.md) first.
## License
MIT
---
**"The difference between a strong model and a strong system is the relentless pursuit of truth."**
This project is in active development toward Level 3. Star the repo and follow along if you want to push Grok Build to its absolute limits.