The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions in SKILL.md direct the agent to run a shell command (bash scripts/brand.sh init ...) using user-provided values for --name and --one-liner. This creates a command injection risk where a user could provide a string containing shell metacharacters (e.g., ;, backticks, or $()) to escape the provided quoting and execute unauthorized commands on the host system.
[COMMAND_EXECUTION]: The bundled script scripts/brand.sh performs unsafe file manipulations. In its set command, it uses the sed utility to update values (sed -E "s|^($key:).*|\\1 $value|") without sanitizing the $value or $key variables. A malicious input containing the sed delimiter or other control characters could lead to arbitrary file modification or script execution within the sed context.
[COMMAND_EXECUTION]: The script lacks input validation for the --out and --register paths, which are passed directly to mkdir -p and used for file redirections. These parameters could be manipulated to write files to unintended locations on the filesystem, bypassing the intended brand folder structure.

brand-init