The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates machine learning tasks by guiding the agent to generate and execute Python scripts via uv run or pixi run. These operations are central to the skill's purpose but involve the execution of shell commands and locally generated code.
[EXTERNAL_DOWNLOADS]: The skill references and fetches machine learning framework dependencies, such as PyTorch wheels, from the official and well-known download.pytorch.org repository.
[PROMPT_INJECTION]: The skill ingests data from user-provided journal files and script-generated logs to create summaries and visualizations, creating a surface for indirect prompt injection. Ingestion points: Reads JOURNAL.md, IGNORED_RUNS.md, and log files in the runs/ directory. Boundary markers: None identified. Capability inventory: Shell command execution (uv run), file creation, and data visualization. Sanitization: Uses regex and JSON parsing to extract metrics from logs, while journal content is included in final report sections.

ml-experimentation