llm-gateway
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFECREDENTIALS_UNSAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The configuration examples for Docker Compose and LiteLLM include hardcoded database credentials ("postgresql://litellm:password@postgres:5432/litellm") in SKILL.md. While the documentation notes these should be replaced, providing functional default credentials in templates can lead to accidental exposure if users do not replace them before deployment.
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the deployment of an API gateway that handles untrusted data from users and third-party LLM providers.
- Ingestion points: The LiteLLM proxy service exposed on port 4000 in SKILL.md, which receives chat and completion requests.
- Boundary markers: Absent; the templates in SKILL.md do not include instructions or delimiters to help underlying models distinguish between system instructions and user-provided content.
- Capability inventory: The gateway routes traffic to external APIs, writes state to PostgreSQL and Redis, and exports operational logs to Langfuse as defined in SKILL.md.
- Sanitization: Absent; no input validation, escaping, or prompt filtering is implemented in the gateway configuration in SKILL.md.
Audit Metadata