The Agent Skills Directory

[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill recommends installing the trackio package via pip. This is the official library from the Gradio organization (https://github.com/gradio-app/trackio), which is a well-known service in the machine learning ecosystem.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill includes functionality to sync training metrics to Hugging Face Spaces and send alerts to external webhooks such as Slack or Discord. These are standard features for experiment monitoring and utilize well-known, trusted services. The documentation provides guidance on data privacy, explaining that Hugging Face Spaces are public by default and demonstrating how to use the private=True flag for sensitive data.
[INDIRECT_PROMPT_INJECTION]: The skill defines a workflow where an AI agent monitors training metrics and alerts to autonomously iterate on experiments. The agent ingests data via CLI commands (trackio list alerts --json, trackio get metric --json) and makes decisions to adjust training code or parameters. While this involves processing external data, it is a primary functional requirement of the tool and is handled as a standard autonomous agent workflow.
[COMMAND_EXECUTION]: The skill uses the trackio CLI to query experiment data. All commands are restricted to the tool's intended metrics management functionality and do not involve suspicious command patterns or privilege escalation.

huggingface-trackio