The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an Indirect Prompt Injection surface where untrusted data is ingested and passed to AI models. \n
Ingestion points: The data_processing.sh script reads local text files using cat, and conditional_workflow.sh accepts input via command-line arguments. \n
Boundary markers: Absent; external content is directly interpolated into the prompt strings without delimiters. \n
Capability inventory: The skill uses the infsh CLI to interact with various AI models (e.g., Claude, Flux) and can perform network requests via curl. \n
Sanitization: Absent; external data is placed raw into the model's JSON input payload. \n- [COMMAND_EXECUTION]: Persistence Patterns. The skill provides detailed examples for using crontab to schedule recurring execution of AI scripts, which allows the code to maintain a persistent presence on the system. \n- [COMMAND_EXECUTION]: Subprocess Execution. The automation.py Python script utilizes the subprocess module to programmatically execute the infsh CLI tool. \n- [DATA_EXFILTRATION]: External Network Patterns. The monitored_workflow.sh script includes a pattern for sending status updates and potential error logs to an external webhook URL using curl.

ai-automation-workflows