The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill directs the user to install official packages from the vendor, specifically the mlflow Python package and the @mlflow/core and @mlflow/openai Node.js packages. These are standard dependencies required for the tracing functionality described.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the capture of application inputs and outputs for observability, which creates a surface for ingesting untrusted data into the MLflow tracing backend.
Ingestion points: Untrusted data such as user messages, LLM payloads, and retrieval results are captured using span.set_inputs() and span.set_outputs() as seen in references/python.md and references/typescript.md.
Boundary markers: The instructions do not prescribe specific delimiters or markers to isolate captured text data, depending instead on the application's internal data structures.
Capability inventory: The skill enables the agent to export captured data over the network to MLflow tracking servers (including Databricks Unity Catalog) and persists trace IDs to the local file system or stdout for verification.
Sanitization: To mitigate the risk of exposing sensitive data, the skill includes a dedicated section and a code recipe in references/advanced-patterns.md for redacting PII and credentials (e.g., password, api_key, token) using span_processors before any data is exported.

mlflow-trace-instrumentation