$ show x1xhlol/system-prompts-and-models-of-ai-tools
This repository is a curated collection of system prompts, tool definitions, and behavioral configurations extracted from dozens of commercial and open-source AI coding assistants and products. It contains over 100 text files organized by vendor (Anthropic, Cursor, VSCode Agent, Manus, Devin, Replit, etc.), each representing a snapshot of how these AI agents are instructed to behave, what tools they can use, and what safety rules they follow. The collection is not an application or a data pipeline; it is a reference archive for researchers, prompt engineers, and developers studying AI agent design patterns, prompt injection risks, and competitive analysis. The repository includes both full system prompts (e.g., Claude Code 2.0, Cursor Agent, Manus) and structured tool JSON schemas that define function-calling APIs. While the content is valuable for understanding AI agent architecture, the repository itself has no executable code, no versioning of individual prompts, and no automated validation, making it a static snapshot rather than a living project.
The LLM identified 4 files worth highlighting for reuse:
A comprehensive JSON file defining 25+ tools for a general-purpose AI agent, including file operations, browser control, shell execution, web search, and deployment. Each tool follows the OpenAI function-calling schema with name, description, and typed parameters. This schema is directly reusable as a template for defining tool sets for any AI agent, and demonstrates a well-structured, complete tool interface pattern.
A detailed system prompt for the Cursor coding agent that defines a structured workflow: analyze task, create todo list, implement changes with status updates, and verify results. It includes explicit rules for code style, parallel tool execution, and error handling. The todo management pattern and status update protocol are reusable templates for any agent that needs to track multi-step tasks.
The full system prompt for Anthropic's Claude Code CLI agent, including tool definitions (Bash, Read, Edit, Write, WebFetch, WebSearch, Task), tone/style rules, proactiveness guidelines, and a complete git safety protocol for creating PRs. The structured XML-tagged configuration and the separation of behavior sections (e.g., <claude_behavior>, <refusal_handling>) provide a reusable template for defining AI agent behavior in a modular way.
A system prompt for Perplexity's web automation agent with extensive security rules, including prompt injection defense layers, social engineering protection, content isolation, and action categorization by risk level. The structured approach to defining prohibited actions, explicit permission categories, and injection detection patterns is a reusable security framework for any AI agent that interacts with untrusted web content.