The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The scripts/connect.py file defines an installation process for the Higgsfield CLI that downloads a shell script from 'https://raw.githubusercontent.com/higgsfield-ai/cli/main/install.sh' and executes it by piping to 'sh'. This method allows for the execution of arbitrary remote code on the user's system.
[COMMAND_EXECUTION]: The scripts extensively use 'subprocess.run()' to call external programs. In scripts/connect.py, the installer commands are executed with 'shell=True', a practice that increases the risk of command injection. scripts/creator.py and scripts/media.py also execute various CLI tools like 'ffmpeg' and several AI provider wrappers, often incorporating user-provided data.
[EXTERNAL_DOWNLOADS]: The skill facilitates downloading and running installers and packages from external repositories, including a GitHub-hosted shell script for Higgsfield and an NPM package for Codex. These downloads come from sources not recognized as trusted organizations, posing a supply chain risk.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied scripts and prompts for audio and image generation. Because these inputs are used to construct commands and API calls, and because the skill has broad system access (shell, network, file system), a malicious payload hidden in a script file or a deceptive API response could lead to indirect prompt injection and unauthorized actions.
[CREDENTIALS_UNSAFE]: The skill handles plaintext API keys for multiple providers (Gemini, OpenAI, ElevenLabs, Azure). Although it attempts to secure the stored keys by using 'chmod 600' on the configuration file, the keys are managed in plaintext within the environment and scripts.

creator-audio