The Agent Skills Directory

[COMMAND_EXECUTION]: Use of osascript to execute shell commands on the host machine from within a virtualized environment (Cowork), which represents a deliberate bypass of environment isolation boundaries.
[COMMAND_EXECUTION]: Unsafe interpolation of the $CONFIG shell variable into a Python script literal in Step 1. If the configuration file content contains triple single quotes or malicious code, it can lead to arbitrary code execution when the script is invoked.
[COMMAND_EXECUTION]: Use of an unquoted heredoc delimiter (JSONEOF) in Step 3, which allows the shell to perform parameter expansion and command substitution on variables like $FILENAME. If a file name contains backticks or $(), the commands will be executed by the shell during payload construction.
[DATA_EXFILTRATION]: The skill reads file contents from the local file system (including the user's Documents folder and session outputs) and transmits them to a remote URL via curl POST requests.
[CREDENTIALS_UNSAFE]: The skill programmatically accesses a private configuration file at ~/.cowork-gdrive-config.json to retrieve a sensitive API endpoint URL, which is a high-risk operation for an automated agent.

google-drive-upload