The Agent Skills Directory

[Indirect Prompt Injection Surface]: The documentation provides an example of a research tool that passes user-provided input directly to a subagent. While this is a common architecture for agentic systems, it creates a surface where the input could influence the subagent's instructions.
Ingestion points: The topic string parameter in the research function within SKILL.md is passed to the subagent.
Boundary markers: The code snippet does not use explicit delimiters or "ignore" instructions to separate the topic from the subagent's system message.
Capability inventory: The tool creates and runs an ai.Agent instance, representing a subagent capability.
Sanitization: There is no explicit sanitization or validation of the topic variable before it is used in ai.user_message(topic). Providing this example without safety markers is a security consideration for developers adopting the pattern.

ai-python-streaming-tools