The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements several patterns that ingest untrusted data and interpolate it directly into AI model prompts without adequate protection.
Ingestion points: data_processing.sh reads local text files using $(cat $file); conditional_workflow.sh accepts user-provided command-line arguments via $1.
Boundary markers: Absent. Input data is directly concatenated into the JSON payload of the prompt.
Capability inventory: The skill has the capability to execute shell commands, perform network operations via curl, and interact with multiple AI models through the belt CLI.
Sanitization: No escaping or validation is performed on the data before it is sent to the models.
[COMMAND_EXECUTION]: The skill relies extensively on executing shell commands and scripts (Bash and Python) that invoke the belt CLI tool to perform AI operations.
[PERSISTENCE]: The skill provides explicit instructions for configuring cron jobs (crontab -e) to schedule recurring execution of automation scripts, allowing the skill's logic to persist and run automatically over time.
[EXTERNAL_DOWNLOADS]: The documentation references external resources and installation instructions hosted on raw.githubusercontent.com and suggests installing the belt-sh/cli package via npx.

ai-automation-workflows