litellm

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The prompt includes explicit secret-like values (e.g., LITELLM_MASTER_KEY=sk-1234 and a DATABASE_URL with user:pass) and demonstrates embedding them directly in Docker/CLI commands and config fields, which would require the agent to output secret values verbatim.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.80). The skill documentation explicitly states LiteLLM will accept and process user-generated requests from Open WebUI (references/integrations/openwebui.md) and route to external model providers and callbacks (SKILL.md and references/*), meaning it ingests untrusted third‑party content (user messages, provider responses, external config/JWKS) that can materially influence routing, callbacks, and downstream actions.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.70). The Quick Start explicitly runs a remote Docker image (docker.litellm.ai/berriai/litellm:main-stable) which is fetched at runtime and executed, so this is a required external dependency that runs remote code.