The Agent Skills Directory

[COMMAND_EXECUTION]: The references/rule.md file contains a Node.js script example (scripts/check-image-optimisation.mjs) that is vulnerable to shell command injection. It uses child_process.execSync with a template literal that directly interpolates file paths into a shell command string: execSync("identify -verbose \"${imgPath}\" ..."). An attacker could trigger arbitrary code execution by including a file with a crafted name containing shell metacharacters (e.g., backticks or semicolons) in the scanned directory.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it processes untrusted data from image assets. 1) Ingestion points: Image filenames and metadata within the project directory. 2) Boundary markers: The instructions lack delimiters or warnings to ignore instructions potentially embedded in image metadata. 3) Capability inventory: The skill utilizes shell-based tools and provides a script capable of command execution. 4) Sanitization: The provided code examples lack proper sanitization for handling untrusted filenames before passing them to the shell.

optimized