The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes templates that process external data from files or command-line arguments and interpolate them into AI model prompts. This creates a surface for indirect prompt injection if the processed content is untrusted.
Ingestion points: Ingestion of $1 in conditional_workflow.sh and $(cat $file) in data_processing.sh.
Boundary markers: Not present in the provided templates.
Capability inventory: The templates use the belt CLI to run AI apps, write to the local filesystem, and use curl for alerting.
Sanitization: No data sanitization or escaping is implemented in the examples.
[EXTERNAL_DOWNLOADS]: The skill references external resources for tool installation and provides commands to add additional modular skills from the vendor's repository.
Evidence: Usage of npx skills add for several inference-sh/skills packages and documentation links to raw.githubusercontent.com/inference-sh/skills/.

ai-automation-workflows