The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references the official Langfuse cloud domain (cloud.langfuse.com) for tracing data. This is a well-known service for LLM observability.
[CREDENTIALS_UNSAFE]: The code examples include configuration for public_key and secret_key using clear placeholders ('pk-...' and 'sk-...'). No actual credentials or secrets are exposed.
[INDIRECT_PROMPT_INJECTION]: The skill demonstrates patterns for ingesting user input into LLM chains (e.g., in LangChain templates). While this represents a common attack surface for LLM applications, the skill follows standard implementation patterns for the libraries discussed.
Ingestion points: Input variables in LangChain ChatPromptTemplate and OpenAI generation calls.
Boundary markers: Uses standard library abstractions like ChatPromptTemplate which provide basic separation.
Capability inventory: Tracing LLM calls, monitoring cost, and logging generations.
Sanitization: Relies on the underlying LLM provider and library defaults.

langfuse