The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads several well-known libraries and benchmarking tools from their official GitHub repositories, such as arcee-ai/mergekit, lm-sys/FastChat, and EleutherAI/lm-evaluation-harness. These are standard tools in the machine learning and natural language processing community.
[COMMAND_EXECUTION]: The skill includes Python scripts that use subprocess.run to call the mergekit-yaml command-line utility. This execution is confined to the intended purpose of automating model merging configurations and is a typical implementation for tool wrappers.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of evaluation datasets and prompts for benchmarking merged models. It relies on established evaluation harnesses which use standard isolation practices to prevent embedded data from overriding the agent's behavior.

model-merging