The Agent Skills Directory

[COMMAND_EXECUTION]: The example calculate tool implementation in reference/adk-core-patterns.md utilizes Python's eval() function to process input expressions. Although the code includes a developer note suggesting replacement with a safer alternative for production, the use of eval() on strings derived from model output creates a direct vulnerability for arbitrary code execution if the agent is manipulated via prompt injection.- [PROMPT_INJECTION]: The skill implements a architecture where agents ingest untrusted user data through FastAPI endpoints (as shown in reference/adk-fastapi-integration.md) and are granted high-privilege capabilities including Bash execution and file system Write and Edit access (as specified in SKILL.md). This combination presents a surface for indirect prompt injection, where an attacker could embed malicious instructions in conversation history or retrieved data to influence the agent to perform unauthorized system or file operations.

google-adk