The Agent Skills Directory

[COMMAND_EXECUTION]: Provides numerous script templates designed to automate tasks using the belt CLI tool. The skill correctly implements the principle of least privilege by using the allowed-tools field to restrict the Bash tool to only execute belt commands.
[EXTERNAL_DOWNLOADS]: Includes links to official installation guides and commands to add related skills from the inference-sh and belt-sh GitHub organizations, which are the legitimate sources for the vendor's tools.
[DATA_EXFILTRATION]: Features a monitoring template that demonstrates sending error alerts to a remote webhook. The script uses https://your-webhook.com/alert as a placeholder, clearly intended for user-specific configuration.
[COMMAND_EXECUTION]: Provides educational examples for setting up task persistence via crontab. These are presented as manual configuration steps for the user rather than automated actions by the agent.
[PROMPT_INJECTION]: The 'Data Processing Pipeline' example demonstrates reading local file content into an LLM prompt. This creates a surface for indirect prompt injection if source files contain adversarial content, which is a noted risk factor for data analysis workflows.
Ingestion points: Local files read from ./data/raw/ in the data_processing.sh template.
Boundary markers: Not present in the educational template.
Capability inventory: The skill uses the belt tool to communicate with remote AI inference services.
Sanitization: None; the example demonstrates basic data interpolation for functional clarity.

ai-automation-workflows