The Agent Skills Directory

[Indirect Prompt Injection] (MEDIUM): The customer_interview_analyzer.py script is designed to process untrusted external content from interview transcripts.
Ingestion points: The script reads from interview_transcript.txt and interview.txt as seen in the usage examples.
Boundary markers: Absent. The documentation does not provide instructions for the agent to use delimiters or ignore embedded instructions within transcripts.
Capability inventory: The script extracts feature requests, sentiment, and themes that directly influence product roadmaps and document creation.
Sanitization: Absent. No filtering or validation logic is mentioned for handling malicious input within transcripts.
[Command Execution] (MEDIUM): The skill documentation explicitly instructs the agent to execute local Python scripts (scripts/rice_prioritizer.py and scripts/customer_interview_analyzer.py).
Evidence: Multiple execution commands are provided in SKILL.md (e.g., python scripts/rice_prioritizer.py features.csv).
Risk: Because the code for these scripts is not included in the skill definition, their actual operations cannot be verified, which is a concern for a toolkit intended to process business data.

product-manager-toolkit