The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands to manage experiment workflows, specifically using cd to navigate directories and uv run python to execute the experiment scripts.
[REMOTE_CODE_EXECUTION]: The agent is instructed to generate Python scripts based on user requirements and then execute them. While this involves running dynamically generated code, it is the primary and stated purpose of the skill for running ML experiments.
[DATA_EXFILTRATION]: The skill integrates with Comet.ml through the liblaf.cherries library to log experiment metrics and assets. This is a standard and transparent functionality for experiment tracking services.
[DATA_EXPOSURE]: The skill accesses local files including logs, data artifacts, and configuration files within the experiment directory structure to generate reports and analyze results.

run-cherries-experiments