The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill includes examples of processing local data files by interpolating their content directly into AI prompts (e.g., in data_processing.sh using $(cat $file)). This creates a surface where malicious content within the processed files could potentially influence agent behavior.
Ingestion point: data_processing.sh reads text files from ./data/raw.
Boundary markers: None are used to delimit the interpolated file content from the surrounding prompt instructions.
Capability inventory: The skill utilizes the belt CLI to execute tasks across various AI models.
Sanitization: No sanitization or validation of the file content is implemented before interpolation.
[EXTERNAL_DOWNLOADS]: The skill documentation references official installation scripts on GitHub (raw.githubusercontent.com/inference-sh) and hosts assets on the service's own domain (cloud.inference.sh). These are legitimate resources associated with the skill's primary functionality.
[DATA_EXFILTRATION]: An example in monitored_workflow.sh demonstrates the use of curl to send error alerts to an external webhook. This is provided as a configuration template (https://your-webhook.com/alert) intended for the user to define their own monitoring endpoint.

ai-automation-workflows