The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends using official packages and tools from the Model Context Protocol ecosystem, such as npx @modelcontextprotocol/create-server, pip install mcp, and @modelcontextprotocol/sdk. These are standard, official resources for MCP development.
[COMMAND_EXECUTION]: Provides standard initialization and testing commands (e.g., npx @modelcontextprotocol/inspector) used for local development and protocol compliance verification.
[INDIRECT_PROMPT_INJECTION]: The skill documents the implementation of tools that process external input, which represents a potential attack surface. However, it includes extensive guidance on mitigation:
Ingestion points: Identified in SKILL.md and references/tools.md as tool arguments and resource contents.
Boundary markers: The use of JSON-RPC 2.0 provides structural boundaries for messages.
Capability inventory: Handlers for file system access, database queries, and HTTP requests are described in references/resources.md and references/tools.md.
Sanitization: The skill provides explicit code examples for input validation using Zod and Pydantic, regex-based SQL injection prevention for table names, and robust path resolution techniques to prevent directory traversal.
[CREDENTIALS_UNSAFE]: The constraints in SKILL.md explicitly forbid hardcoding credentials and secrets, recommending the use of environment variables instead.

mcp-developer