The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill's primary function is to facilitate the execution of user-defined Python code on remote managed infrastructure (CPUs, GPUs, and TPUs). This is achieved via the hf_jobs tool, which accepts script content as strings or URLs to be executed within remote containers.
[COMMAND_EXECUTION]: Instructions guide the agent to perform job management tasks such as submitting, inspecting, monitoring, and canceling remote processes. The skill utilizes the uv and run execution modes for containerized workloads.
[EXTERNAL_DOWNLOADS]: The skill fetches configuration and scripts from Hugging Face's official GitHub and Hub repositories (e.g., huggingface/trl, uv-scripts organization). These operations target a well-known technology provider and follow standard integration patterns.
[DATA_EXFILTRATION]: The skill handles sensitive authentication tokens (HF_TOKEN). It implements secure management by promoting the use of the platform's automatic secret substitution feature ($HF_TOKEN), which ensures tokens are handled server-side and are not exposed in logs or code during the submission process.
[PROMPT_INJECTION]: The skill contains surfaces for indirect prompt injection (Category 8) as it processes data from external Hub repositories. The provided scripts utilize standard chat templates and role-based structures to provide separation between data and instructions. The risk is minimized by the sandboxed nature of the job environment.

hugging-face-jobs