The Agent Skills Directory

[Indirect Prompt Injection] (MEDIUM): The skill processes untrusted external text and file content without sanitization or boundary markers.
Ingestion points: The truncate_text function in scripts/text_truncator.py accepts raw strings, and truncate_text_from_file accepts file paths.
Boundary markers: Absent. The skill does not implement delimiters or 'ignore' instructions for the data it processes.
Capability inventory: The script has the ability to read local files via open(). While it cannot write or execute commands, the output is returned to the agent context.
Sanitization: None. The text is truncated purely based on length and punctuation.
[Data Exposure & Exfiltration] (MEDIUM): The truncate_text_from_file method in scripts/text_truncator.py allows reading any file on the system that the agent process has permissions for.
Evidence: Line 139 in scripts/text_truncator.py: with open(file_path, 'r', encoding=encoding) as f:.
Risk: If an agent is manipulated via prompt injection to provide sensitive paths (e.g., SSH keys or environment files) to this tool, the content will be exposed within the agent's reasoning chain or conversation history.

text-truncator