The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The file scripts/connect.py contains a command to download and execute a shell script from a remote URL (https://raw.githubusercontent.com/higgsfield-ai/cli/main/install.sh) by piping it directly to sh. This pattern allows for the execution of arbitrary code from an untrusted external source without prior validation or integrity checks.
[COMMAND_EXECUTION]: The script scripts/connect.py executes system commands using subprocess.run() with shell=True. This is specifically used to run the installation string for the Higgsfield CLI, which is a high-risk pattern that increases vulnerability to command injection if variables are improperly handled.
[EXTERNAL_DOWNLOADS]: The skill depends on several third-party command-line tools and packages that are downloaded during the connection process. While some sources are trusted, others such as the Higgsfield and Antigravity CLI tools originate from third-party domains outside of the primary trusted organization scope.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user data, including video titles and headline text, which are interpolated into prompts for AI models and browser-based actions in scripts/creator.py. The lack of explicit boundary markers or input sanitization creates a surface where malicious user data could manipulate the behavior of the underlying AI agents or tools.
[DYNAMIC_EXECUTION]: The skill dynamically constructs and executes media processing commands and API calls based on runtime logic and provider selection. It also includes manual playbooks for browser-based 'computer use' on external platforms like Google AI Studio and Flow, which could be exploited if the inputs driving these actions are compromised.

creator-thumbnail