The Agent Skills Directory

[SAFE]: The skill functions as a static documentation and reference guide for LLM security best practices.
[SAFE]: While the skill contains common prompt injection strings (e.g., 'Ignore all previous instructions'), these are explicitly labeled as examples of attacks within a defensive context to teach the agent or user how to identify and prevent them.
[SAFE]: The provided PHP code snippets for input and output validation are templates for developer use and are not executed by the agent environment.
[SAFE]: The skill correctly identifies and warns against high-risk anti-patterns such as using eval(), piping raw LLM output to shells, or granting excessive permissions to agents.

llm-security