How william-yeh/system_prompts_leaks works
Project Type and Purpose
• This is not a typical full-stack application. Based on the file names (e.g., “chatgpt-4o-image-safety-policies.md”, “claude-3.7-sonnet-full-system-message-humanreadable.md”), this repository is a documentation/knowledge-base of AI system messages, policy guidelines, prompt injections, and persona definitions.
• The README confirms it’s a personal collection of proofs and methods, with no runnable code or services.
File Structure
• All files live at the top level; there is no “src/”, “frontend/”, or “backend/” directory.
• File name patterns cluster around specific AI models or topics (ChatGPT-4, Claude-3.7, Grok, Gemini).
• There are Markdown (.md) and text (.txt or .xml) files—no build scripts, no configuration.
README for Additional Insights
• The README is a short note asking for proof-technique ideas and provides contact channels.
• There is no mention of architecture, dependencies, or runtime environment.
a. Main Components
• File Storage (the Git repository holding all .md/.txt/.xml)
• Category Index (how documents are grouped by model & topic)
• Version Control (Git)
• Authoring/Editing Tool (e.g., VS Code, text editor)
b. Relationships & Interactions
• Users (authors/reviewers) → edit documents in File Storage via Editor
• Git ↔ File Storage for commits, branching, change history
c. Architectural Patterns & Principles
• Single source of truth: the Git repo contains all documents
• Separation of concerns: storage layer (files) vs. presentation layer (editor or future web UI)
• Classification/Taxonomy: grouping docs by model and topic for maintainability
d. Key Technologies & Libraries
• Markdown as the content format
• Git for version control
• Editor or IDE of choice for authoring
• (Optional) Static-site generator (e.g., MkDocs, Docusaurus) or search index (e.g., Elasticsearch)
Elements
• Clear Labels: “Git Repo”, “Markdown Files”, “Editor”, “Search Service (optional)”, “Users”
– User → Editor → Git Repo (commit)
– Search Service ↔ Git Repo (indexing & querying)
Detailed, Essential Architectural Elements
• classification layers: top-level per-model grouping of documents
• authoring workflow: write → stage → commit → review
• If future extensibility is planned, add placeholders for deployment (e.g., Netlify, GitHub Pages) and search integration