The Agent Skills Directory

[PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it ingests untrusted metadata from model registries and external experiments.
Ingestion points: Model metadata, parameters, and run notes retrieved from MLflow or other tracking servers.
Boundary markers: Absent; the instructions do not specify delimiters for data interpolated into model card templates.
Capability inventory: The agent has access to Bash(cmd:*), Write, and Edit tools, allowing it to execute commands or modify files based on instructions potentially embedded in metadata.
Sanitization: No sanitization or validation of the ingested strings is performed before they are used to generate Markdown documentation.
[COMMAND_EXECUTION]: The skill instructions direct the agent to execute shell commands for experiment management and to start network-accessible services, such as an MLflow tracking server (mlflow server --host 0.0.0.0).
[EXTERNAL_DOWNLOADS]: The skill documentation references the installation of common Python packages (mlflow, pandas) and interactions with established cloud service providers (AWS S3 and Google GCS) for artifact storage.
[NO_CODE]: The file scripts/version_control.sh is misleadingly named and described. While its metadata claims it is a Bash script for Git/VCS automation, it actually contains Python code that performs basic file size auditing and JSON validation.

tracking-model-versions