The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The skill processes untrusted external data for analysis and transformation, which is an inherent risk for indirect prompt injection.\n
Ingestion points: scripts/analyze.py and scripts/transform.py read all input data from standard input.\n
Boundary markers: No boundary markers or delimiters are used in the prompt or processing scripts to isolate untrusted content from agent instructions.\n
Capability inventory: The skill is restricted to information processing and transformation; it cannot access the network, write to the file system, or execute arbitrary system commands.\n
Sanitization: Input data is parsed as JSON, CSV, or text without filtering for malicious instructional content or prompt injection patterns.

data-processor