The Agent Skills Directory

[PERSISTENCE_MECHANISMS]: The skill includes a template for an auto-setup script in references/testing-distribution.md that modifies configuration files for Claude Code, Claude Desktop, and Cursor (e.g., ~/.claude/settings.json and ~/Library/Application Support/Claude/claude_desktop_config.json). This is intended to register the MCP server so it remains available to the agent across sessions, but it involves automated writing to application configuration directories.
[INDIRECT_PROMPT_INJECTION]: The skill outlines how to build tools that ingest data from external sources such as web APIs (fetch) and local project files (readFile). It includes security guidance on using Zod schemas for input validation to reduce the risk of the agent executing malicious instructions embedded in that data.
[DATA_EXPOSURE]: The documentation provides best practices for managing sensitive information like API keys, recommending the use of environment variables (process.env) rather than hardcoding credentials directly into the server source code.
[COMMAND_EXECUTION]: The skill provides instructions for standard development workflows using tools like bun, tsc, and npx to initialize projects, install dependencies, and run the server.

mcp-server-dev