The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The skill possesses a data ingestion surface (reading procedures from documentation) and command execution capabilities (run_terminal_cmd). This creates a potential surface where untrusted input in a documentation file could influence subsequent automated actions. Evidence: DOCUMENT and FOLLOW methods in CARD.yml.
[Dynamic Execution] (LOW): The skill provides methods to generate and enhance Python CLI scripts based on textual procedures. While the provided examples use standard argparse templates, the capability to write executable code to the filesystem is a minor concern. Evidence: EXTRACT and ENHANCE methods in CARD.yml.
[Command Execution] (LOW): The skill explicitly requests the run_terminal_cmd tool. Although no dangerous commands are hardcoded in the skill itself, this provides the capability to execute arbitrary shell commands on the host environment.

sister-script