The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains phrases like "Ignore previous instructions" and "Pretend you are DAN". These were evaluated and determined to be benign as they are used specifically as examples of malicious patterns that the NeMo Guardrails framework is designed to detect and block. They are not instructions directed at the agent itself.
[EXTERNAL_DOWNLOADS]: The skill instructs the user to install the nemoguardrails Python package and provides links to the official NVIDIA NeMo Guardrails GitHub repository. These are well-known and trusted resources consistent with the skill's stated purpose.
[INDIRECT_PROMPT_INJECTION]: While the skill defines a runtime environment that processes user input, its core functionality is to provide sanitization and safety rails. The implementation uses Colang DSL to define strict boundaries for LLM behavior, which is a defensive security measure.
[DYNAMIC_EXECUTION]: The skill uses RailsConfig.from_content to define safety logic. This is a standard feature of the NeMo Guardrails library for defining programmable rails and does not involve the unsafe execution of arbitrary code from untrusted sources.

nemo-guardrails