The Agent Skills Directory

[DATA_EXFILTRATION]: The templates/file_processor.py file contains tools that allow reading arbitrary files from the filesystem via a user-provided path. The implementation using expanduser() enables access to the home directory, which could facilitate the exposure of sensitive system configuration or credential files if the agent is directed to access them.
[COMMAND_EXECUTION]: The templates/database_server.py file provides a tool for executing raw SQL queries on a SQLite database. Although it includes safeguards such as a read-only connection and a basic validation for SELECT statements, dynamic SQL execution represents a surface for unauthorized data access.
[PROMPT_INJECTION]: The templates/file_processor.py template creates an indirect prompt injection surface by ingesting untrusted content from the local filesystem and providing it to the agent context.
Ingestion points: The _read_text function in templates/file_processor.py reads data directly from the local disk.
Boundary markers: No explicit delimiters or boundary instructions are included in the template to isolate file content from agent instructions.
Capability inventory: The tools summarize_text_file, search_text_file, and read_file_resource enable the agent to read and process file contents.
Sanitization: The template does not perform validation or sanitization on the content read from files before returning it to the caller.
[EXTERNAL_DOWNLOADS]: The skill instructions specify the installation of the fastmcp and httpx Python packages from standard registries. Additionally, the api_wrapper.py template is designed to facilitate network requests to external API endpoints as part of its core functionality.

fastmcp