The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple shell commands (magick, rsvg-convert, cp, rm) to process images and manage filesystem assets during icon generation.\n- [COMMAND_EXECUTION]: Command Injection Risk: In Step 5, the skill executes shell commands using the $1 placeholder, which represents the user-supplied source image path. Because this input is interpolated directly into shell strings, an attacker could provide a path containing shell metacharacters (e.g., backticks, $(...), or ;) to execute arbitrary commands on the system.\n- [PROMPT_INJECTION]: Indirect Prompt Injection Surface: The skill automatically extracts the application name from project files in Step 3 to use in later stages of execution.\n
Ingestion points: Files site.webmanifest, package.json, and config/application.rb are read in Step 3.\n
Boundary markers: Absent. The data is extracted and used directly in templates without delimiters or 'ignore' instructions.\n
Capability inventory: The extracted data is written into site.webmanifest (Step 6) and used to modify HTML layout files (Step 7).\n
Sanitization: Absent. There is no evidence of validation or escaping for the strings extracted from these external files.\n- [COMMAND_EXECUTION]: The skill provides instructions for the user to execute commands with elevated privileges (sudo apt install), which promotes insecure practices if the user lacks the expertise to verify the source.

favicon