langfuse

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.85). The skill includes code examples that pass secret_key/public_key values directly as parameters (e.g., secret_key="sk-..."), which encourages embedding secrets verbatim in generated code or outputs and therefore creates an exfiltration risk.

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

• Third-party content exposure detected (medium risk: 0.30). The Langfuse skill’s runtime workflow can ingest outsider-authored free text because it traces and logs LLM inputs/outputs (e.g., user chat messages) via Langfuse during application execution (e.g., generation(... input={"messages": ... "content": ...}) and evaluate_response(question, response)), so an external user can supply the monitored text.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill sets the Langfuse host to https://cloud.langfuse.com (used in client/callback initialization) and calls langfuse.get_prompt at runtime to fetch prompts that directly control model inputs, so the external URL is used to supply runtime prompt content.