The Agent Skills Directory

Indirect Prompt Injection (LOW): The examples in references/examples.md demonstrate direct interpolation of untrusted data into prompts using f-strings (e.g., f"Extract: {text}") without boundary markers or sanitization. While typical for educational content, this establishes an attack surface where malicious input could influence agent behavior. (1) Ingestion points: text and texts variables in references/examples.md. (2) Boundary markers: Absent. (3) Capability inventory: Data extraction and structured analysis. (4) Sanitization: Absent.
Metadata Anomaly (INFO): The code samples consistently reference a non-existent future model claude-sonnet-4-5-20250929. While not explicitly malicious, this is noted as deceptive or inaccurate metadata.

instructor