The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes multiple bash blocks to manage environment state, verify prerequisites, and perform configuration updates. This includes executing local binaries such as rmbc-update-check and rmbc-workspace located within the skill's directory structure.
[COMMAND_EXECUTION]: A critical command injection vulnerability is present in the workspace activation protocol. The instruction executes /bin/rmbc-workspace active "<user's answer>" using unvalidated input from the user. An attacker could provide input containing shell metacharacters (e.g., ;, &&, ||) or command substitution sequences (e.g., $(...), `...`) to execute arbitrary shell commands on the host system.
[REMOTE_CODE_EXECUTION]: The skill dynamically calculates a root directory (_RMBC_ROOT) and executes binaries found within its bin/ subdirectory. Executing code from paths determined at runtime is a risk, as it could be exploited if an attacker can influence the filesystem structure or the variables used to locate these binaries.
[DATA_EXFILTRATION]: The skill contains a telemetry component that executes a local binary (rmbc-analytics) to log usage data including the skill name and active product. While the preamble claims no code or prompts leave the machine, this represents an automated external logging mechanism.
[EXTERNAL_DOWNLOADS]: The skill invokes the open command to launch an external YouTube URL (https://www.youtube.com/watch?v=zI8tNfefH1M) in the user's browser.

offer-stack