langsmith-trace

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

• Third-party content exposure detected (medium risk: 0.30). The skill instructs the agent to query and read user/application execution traces from LangSmith via CLI commands, which can contain arbitrary untrusted user inputs and LLM outputs.

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

• Potentially malicious external URL detected (medium risk: 0.30). The setup instructions download and execute a remote install script from raw.githubusercontent.com for the LangChain CLI tool, which is an unauthenticated external runtime dependency from a third-party source.