The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The script customer_interview_analyzer.py is designed to process interview transcripts from external text files. If these files contain adversarial instructions, they could potentially influence the agent's summary or behavior.
Ingestion points: The main function in scripts/customer_interview_analyzer.py reads a file path from command-line arguments and reads its full content.
Boundary markers: Absent; there are no delimiters or instructions to the LLM to ignore potentially malicious content within the transcript.
Capability inventory: The script extracts insights like pain points and sentiment but does not perform network or privileged system operations.
Sanitization: Absent; the text is analyzed using regex without filtering for embedded command sequences.
[INDIRECT_PROMPT_INJECTION]: The rice_prioritizer.py script reads feature lists from CSV files. Data within these files (e.g., in the 'description' field) could contain instructions that target the agent.
Ingestion points: The load_features_from_csv function in scripts/rice_prioritizer.py reads data from user-provided CSV files.
Boundary markers: Absent.
Capability inventory: The script can write a sample file to disk (sample_features.csv) but has no network access.
Sanitization: Absent; input strings are not sanitized for injection patterns.
[COMMAND_EXECUTION]: The skill instructions facilitate the execution of local Python scripts provided with the skill. While these specific scripts appear benign, the pattern involves executing code that processes external input files.

product-manager-toolkit