cloud-api-integration
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: Documentation in
references/security-examples.mdincludes examples of jailbreak and prompt injection strings (e.g., 'DAN mode', 'ignore all instructions'). These are explicitly identified as test cases within the security documentation intended for validating injection filters, rather than attempts to manipulate the agent. - [EXTERNAL_DOWNLOADS]: The skill recommends using official packages from well-known providers (Anthropic, OpenAI, Google) and common utility libraries (e.g.,
httpx,tenacity). References are for standard development workflows and correspond to trusted organizations. - [CREDENTIALS_UNSAFE]: The skill demonstrates the risks of hardcoded credentials and provides implementation code for secure loading from environment variables or secret managers using
Pydantic.SecretStr. All keys shown in the text are illustrative placeholders. - [INDIRECT_PROMPT_INJECTION]: The implementation patterns explicitly address the risk of untrusted data from external sources and provide multiple layers of defense, including system-level instructions, boundary markers, and output filtering in the code samples.
Audit Metadata