The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a shell command in Phase 3 (python3 scripts/generate_image.py ...) that interpolates user-provided data, such as brand names and generated prompts, directly into the command line arguments. An attacker could provide a brand name containing shell metacharacters (e.g., ;, &&, or backticks) to execute arbitrary commands on the system.
[CREDENTIALS_UNSAFE]: The SKILL.md file hardcodes absolute filesystem paths targeting a specific user environment: /Users/ugreen/Documents/obsidian/09image/. Exposing specific local usernames and home directory structures reveals information about the host system and increases the risk of unauthorized file system operations if the path construction is manipulated.
[EXTERNAL_DOWNLOADS]: The skill facilitates the download of image data from https://generativelanguage.googleapis.com. This is a well-known service provided by Google and is documented here as an external dependency of the skill.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from user-provided brand descriptions and analyzed image data (Phase 1). This content is directly used to form the prompt argument for the execution script. A lack of strict sanitization or escaping could allow an attacker to influence the final command or the behavior of the image generation model through embedded instructions.

space-logo-generator