The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple shell commands and local binaries (e.g., rmbc-update-check, rmbc-analytics, rmbc-workspace) during initialization and post-execution. These binaries are located in paths derived at runtime from environment variables and user home directories (e.g., $_RMBC_ROOT/bin/).
[COMMAND_EXECUTION]: Potential command injection vulnerability in the post-execution workspace setup. The agent is instructed to run /bin/rmbc-workspace active "<user's answer>" where <user's answer> is freeform text provided by the user via AskUserQuestion. If the platform does not sanitize this input, it allows execution of arbitrary shell commands.
[DATA_EXFILTRATION]: The skill logs usage data including the skill name and the active product name to an external service via a background process (rmbc-analytics). Although the user is prompted for telemetry opt-in, the mechanism exfiltrates local environment metadata to a remote server.
[EXTERNAL_DOWNLOADS]: The skill directs the user to an external YouTube video and references an "Inline upgrade flow" that involves reading and following instructions from a rmbc-upgrade skill, which is a common pattern for fetching and executing updated remote code.
[CREDENTIALS_UNSAFE]: The skill directly reads and modifies a local configuration file at ~/.rmbc-skills/config.yaml using grep and sed to manage state and settings.

thank-you-page