The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the mlflow CLI and jq utility within bash scripts (scripts/discover_schema.sh and scripts/inspect_turn.sh) to perform trace analysis.
[EXTERNAL_DOWNLOADS]: The mlflow CLI commands fetch session and trace data from a remote MLflow tracking server.
[PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks because it processes and analyzes multi-turn chat logs retrieved from an external MLflow server. These logs contain untrusted user and assistant content that could contain instructions designed to manipulate the agent's behavior during analysis.
Ingestion points: Data is ingested via mlflow traces search and mlflow traces get commands and stored in local JSON files.
Boundary markers: None present; the agent is instructed to read and process JSON fields directly.
Capability inventory: Shell execution (mlflow, jq, cat, etc.) and file system access.
Sanitization: No sanitization of trace attributes, inputs, or outputs is performed before processing.

analyzing-mlflow-session