The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes uv run python to execute local scripts (ML.scripts.retrain) and system commands (ls, ps) for environment checks and process monitoring. These operations are scoped to the project directory and are essential for the training workflow.
[PROMPT_INJECTION]: There is a potential surface for indirect prompt injection as the agent is instructed to read generated reports (ML/artifacts/reports/retrain_report_*.md). These reports are derived from external training data. However, this is a standard reporting function and no explicit injection patterns were detected.
[DATA_EXFILTRATION]: The skill operates entirely on local file paths (../data/, ML/artifacts/). There are no commands that perform network requests or attempt to exfiltrate sensitive information.

ml-retrain