The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains instructions that reduce user oversight. Specifically, the directive 'Act first, explain briefly. Don't over-ask for confirmation on routine work' encourages the agent to operate autonomously, which could be exploited to execute unauthorized actions before the user can intervene.
[COMMAND_EXECUTION]: The skill is explicitly designed to perform system automation, including writing and executing PowerShell scripts (.ps1), managing system services, and creating scheduled tasks. While these are stated capabilities, they represent a significant control surface over the host machine.
[DATA_EXFILTRATION]: The skill documentation reveals internal infrastructure details, including specific IP addresses (192.168.1.129), UNC network paths (\\192.168.1.129\SkippyKB), and detailed hardware specifications of the host machine ('My-Server').
[REMOTE_CODE_EXECUTION]: The orchestration capabilities involve processing data from external sources, such as 'Knowledge harvesting & web scraping' via the 'Cassian' persona. This creates a surface for indirect prompt injection where malicious content from the web could influence the agent's high-privilege automation tasks.

rodimus