https://github.com/shuyhere/system-prompt-observatory
Track and compare system prompts across all major AI agents. Diff viewer + component analysis + evolution dashboard.
https://github.com/shuyhere/system-prompt-observatory
ai-agents chatgpt claude-code diff-viewer llm prompt-engineering system-prompts
Last synced: 21 days ago
JSON representation
Track and compare system prompts across all major AI agents. Diff viewer + component analysis + evolution dashboard.
- Host: GitHub
- URL: https://github.com/shuyhere/system-prompt-observatory
- Owner: shuyhere
- Created: 2026-04-06T11:38:02.000Z (21 days ago)
- Default Branch: main
- Last Pushed: 2026-04-06T11:48:33.000Z (21 days ago)
- Last Synced: 2026-04-06T13:28:59.578Z (21 days ago)
- Topics: ai-agents, chatgpt, claude-code, diff-viewer, llm, prompt-engineering, system-prompts
- Language: TypeScript
- Homepage: https://shuyhere.github.io/system-prompt-observatory/
- Size: 3.59 MB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# System Prompt Observatory
Track and compare system prompts across all major AI agents. Git-style diffs powered by Monaco Editor, with component analysis and evolution tracking.
## Live Site
Deployed via GitHub Pages.
## Agents Tracked (670 prompt files)
| Agent | Provider | Versions |
|-------|----------|----------|
| Claude Code | Anthropic | 275 |
| Gemini CLI | Google | 133 |
| Codex CLI | OpenAI | 53 |
| ChatGPT | OpenAI | 46 |
| Claude.ai | Anthropic | 32 |
| Kimi CLI | Moonshot | 32 |
| Other Agents | Various | 29 |
| OpenHands | Open Source | 22 |
| Grok | xAI | 14 |
| Gemini Web | Google | 13 |
| Cursor | Cursor | 11 |
| Perplexity | Perplexity | 10 |
## Features
- **Diff Viewer** — Monaco-powered side-by-side diffs between any two versions
- **Component Analysis** — Token-counted breakdown into 16 categories (Identity, Safety, Tools, Memory, etc.)
- **Evolution Dashboard** — Claude Code prompt size, churn, and composition over 275 versions
- **Multi-source** — Data from cchistory, system_prompts_leaks, jujumilk3, LouisShark, git histories
- **Dark/Light mode**
## Development
```bash
cd web
npm install
npx vite dev # dev server on :3000
npx vite build # production build to dist/
```
## Data Sources
- [cchistory](https://github.com/badlogic/cchistory) — Claude Code versions
- [system_prompts_leaks](https://github.com/asgeirtj/system_prompts_leaks) — Multi-agent leaks
- [jujumilk3/leaked-system-prompts](https://github.com/jujumilk3/leaked-system-prompts) — Historical captures
- [LouisShark/chatgpt_system_prompt](https://github.com/LouisShark/chatgpt_system_prompt) — ChatGPT history
- [openai/codex](https://github.com/openai/codex) — Codex CLI git history
- [google-gemini/gemini-cli](https://github.com/google-gemini/gemini-cli) — Gemini CLI git history
- [MoonshotAI/kimi-cli](https://github.com/MoonshotAI/kimi-cli) — Kimi CLI git history
- [All-Hands-AI/OpenHands](https://github.com/All-Hands-AI/OpenHands) — OpenHands git history
## Inspired By
- [cchistory.mariozechner.at](https://cchistory.mariozechner.at)
- [How System Prompts Define Agent Behavior](https://www.dbreunig.com/2026/02/10/system-prompts-define-the-agent-as-much-as-the-model.html) by Drew Breunig
- [nilenso/context-viewer](https://github.com/nilenso/context-viewer)
## License
Data sourced from public repositories under their respective licenses. Site code is MIT.