The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external model artifacts, infrastructure configurations, and monitoring data, which serves as a potential entry point for indirect prompt injection attacks.
Ingestion points: The skill explicitly accesses model artifacts, infrastructure configs, and monitoring data as part of its deployment and optimization tasks.
Boundary markers: There are no explicit instructions or delimiters provided to distinguish between system instructions and data from these external files.
Capability inventory: The skill utilizes Bash, Write/Edit, Read, and Glob/Grep tools, providing a wide range of actions that could be triggered by injected instructions.
Sanitization: The skill's documentation includes best practices such as "Input Validation: Sanitize all prediction inputs," though it does not provide specific implementation details for the agent's own internal processing of configuration files.
[COMMAND_EXECUTION]: The skill uses the Bash tool to execute deployment commands, monitoring setups, and performance tests. While these actions are required for the machine learning engineer persona, the execution of arbitrary shell commands represents a significant capability that requires oversight.

machine-learning-engineer