The Agent Skills Directory

[COMMAND_EXECUTION]: The preamble script executes several shell commands and binaries (rmbc-update-check, rmbc-analytics) from directories determined at runtime. These tools appear to be part of the RMBC framework's management suite.
[COMMAND_EXECUTION]: The skill includes a step that runs a shell command using /bin/rmbc-workspace active "<user's answer>". Since this includes freeform input from the user without evident sanitization, it creates a risk of command injection if the user provides malicious shell metacharacters.
[PROMPT_INJECTION]: The skill employs a conditional instruction flow where it may read and follow instructions from a separate file (skills/rmbc-upgrade/SKILL.md) based on the output of an update check. This dynamic loading can bypass static analysis of the skill's behavior.
[EXTERNAL_DOWNLOADS]: The skill references a YouTube video and performs an automated check for updates, which likely involves network communication to fetch version information.
[DATA_EXFILTRATION]: Usage statistics, such as the skill name and product context, are collected and processed by a local analytics binary (rmbc-analytics).

guarantee-writer