The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes building tools that process untrusted data (file paths). It provides high-quality security documentation and code examples (using path resolution and containment checks) to mitigate path traversal risks. Ingestion points include the path parameter in tool handlers, and the skill suggests using the roots/list capability as a boundary marker for allowed directories.
[COMMAND_EXECUTION]: The documentation covers process execution for MCP servers. It explicitly warns against using shell execution and provides examples of secure command invocation using array-based arguments (e.g., execFile instead of exec) to prevent command injection.
[EXTERNAL_DOWNLOADS]: The skill references packaging tools from the @anthropic-ai scope and manifest schemas hosted in the anthropics organization, which are trusted sources. It also mentions standard development utilities like the MCP inspector.

build-mcpb