The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of the trackio Python package from PyPI. This package is maintained by the Gradio team and is documented by Hugging Face, which is a trusted organization in the machine learning ecosystem.
[COMMAND_EXECUTION]: The skill utilizes the trackio CLI tool for querying project metadata, run summaries, and experiment metrics. These commands are executed locally as part of the intended training monitoring workflow.
[DATA_EXFILTRATION]: The skill supports sending diagnostic alerts to external webhook URLs (e.g., Slack, Discord) via the trackio.alert() function and related configuration. While this is a standard feature for real-time monitoring, it represents a path for training diagnostics to be sent to external endpoints.
[PROMPT_INJECTION]: The skill implements an autonomous workflow where an agent monitors training progress via alerts to drive iteration. This creates a surface for indirect prompt injection: 1. Ingestion points: Alert titles and descriptions are retrieved from the experiment tracking database via trackio list alerts --json (references/alerts.md). 2. Boundary markers: The documentation does not specify delimiters or instructions to ignore embedded commands within alert text. 3. Capability inventory: The agent has the capability to stop training runs, adjust hyperparameters, and launch new training processes based on the alert content. 4. Sanitization: No sanitization or validation of the alert data is described before the agent processes it to make autonomous decisions.

huggingface-trackio