The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands like curl, wget, and node using parameters derived from external data such as image_url in thread.json. This pattern is vulnerable to command injection if the agent does not strictly sanitize the input strings before execution.
[EXTERNAL_DOWNLOADS]: The workflow involves searching for and downloading images, logos, and spritesheets from arbitrary internet sources found via WebSearch or provided in user data files. This introduces risks of processing malicious files or being directed to attacker-controlled servers.
[REMOTE_CODE_EXECUTION]: Downloaded assets are subsequently processed by local scripts such as build-character.mjs. This multi-step chain where untrusted remote data is processed by local tools could lead to code execution if vulnerabilities exist in the processing scripts or their dependencies like face-api.js.
[PROMPT_INJECTION]: The skill processes untrusted data from thread.json and instructs the agent to "study" external images for character design. This represents an indirect prompt injection surface where malicious metadata or hidden text in images could influence the agent's behavior during the asset generation process.

game-assets