litellm-guide
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill provides structured guidance for LiteLLM integration and development. It follows secure coding practices by recommending the use of environment variables (e.g.,
os.environ/OPENAI_API_KEY) instead of hardcoding secrets. - [INDIRECT_PROMPT_INJECTION]: The skill defines workflows for processing data from external LLM providers and user inputs, which presents a surface for indirect injection.
- Ingestion points: External data enters through user prompts, provider responses, and logs processed by the LiteLLM Proxy or SDK (documented in
references/proxy-gateway.mdandreferences/sdk-api.md). - Boundary markers: The
SKILL.mdfile explicitly instructs the agent to treat model responses, MCP tool outputs, and logs as untrusted input rather than instructions. - Capability inventory: The skill describes capabilities including network operations to LLM providers, database interactions via Prisma, and tool execution through the Model Context Protocol (MCP).
- Sanitization: Guidance is provided on using built-in guardrails, PII masking, and redaction features to sanitize outgoing and incoming data.
Audit Metadata