The Agent Skills Directory

[DATA_EXFILTRATION]: The scripts/rice_prioritizer.py script accepts a file path as a command-line argument and opens it for reading in the load_features_from_csv function without performing path validation or boundary checks. Unlike the interview analyzer script in the same toolkit, this script does not resolve or sanitize the input path, potentially allowing an agent to be manipulated into reading sensitive system files if prompt injection occurs.
[PROMPT_INJECTION]: The scripts/customer_interview_analyzer.py script processes untrusted text data from customer interview transcripts and uses regular expressions to extract quotes, themes, and feature requests. This ingestion of external data creates an attack surface for indirect prompt injection. Maliciously crafted instructions hidden within a transcript (e.g., in HTML comments or specific text patterns) could influence the agent's behavior when it interprets the extracted results, as the script lacks boundary markers or sanitization to distinguish between data and instructions.

product-manager-toolkit-skill