The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches the langfuse-mcp package from the public PyPI registry using the uvx tool during the setup process described in SKILL.md and references/setup.md.
[DATA_EXFILTRATION]: The skill interacts with the Langfuse API (defaulting to https://cloud.langfuse.com) to exchange observability data such as LLM traces and prompt configurations. This involves the use of sensitive user-provided API keys (LANGFUSE_PUBLIC_KEY and LANGFUSE_SECRET_KEY). As Langfuse is a well-known service, this communication is consistent with the skill's intended functionality.
[COMMAND_EXECUTION]: The installation process involves executing shell commands to configure the MCP server and manage environment variables as outlined in the setup instructions.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface because it processes external data (traces, observations, and prompt content) from the Langfuse API.
Ingestion points: Data entering through tools such as fetch_trace, fetch_observation, and get_prompt in references/tool-reference.md.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are included in the tool documentation or examples.
Capability inventory: The skill includes tools to create or update prompts and datasets, such as create_text_prompt and create_dataset_item listed in references/tool-reference.md.
Sanitization: There is no documentation regarding the validation or sanitization of data retrieved from the Langfuse API.

langfuse