The Agent Skills Directory

[SAFE]: No security findings or malicious patterns were identified across the skill files.
[DATA_EXFILTRATION]: The skill does not perform any network operations, nor does it access sensitive local file paths or hardcoded credentials.
[REMOTE_CODE_EXECUTION]: No patterns involving remote script downloads, piped execution, or installation of untrusted dependencies were found.
[COMMAND_EXECUTION]: The provided Python script (scripts/process.py) is limited to text processing and mathematical scoring; it does not utilize subprocess, os.system, or any other shell execution mechanisms.
[PROMPT_INJECTION]: While the skill contains instructional content regarding prompt injection as a risk category for LLMs, it does not include instructions designed to bypass the agent's own safety guardrails or override its behavior.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted text data through its PII scanner. However, the analysis confirms that there are no dangerous sinks (such as dynamic execution or shell commands) that could be triggered by malicious input within the processed text.
[DYNAMIC_EXECUTION]: The scripts do not use eval(), exec(), or dynamic module loading. The logic is strictly based on regular expressions and deterministic scoring algorithms.

llm-output-privacy-risk