The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The scripts/connect.py script facilitates the installation of the Higgsfield CLI by downloading a shell script from a remote repository and piping it directly to the system shell (sh). This pattern is a significant security risk as it executes unverified code from an external source.
Evidence: curl -fsSL https://raw.githubusercontent.com/higgsfield-ai/cli/main/install.sh | sh is defined as an installation command and executed using subprocess.run(..., shell=True).
[COMMAND_EXECUTION]: The skill's core scripts (connect.py, creator.py, and media.py) rely heavily on executing external binaries and shell commands to perform their tasks.
Evidence: Frequent use of subprocess.run to invoke ffmpeg, ffprobe, gcloud, npm, higgsfield, agy, and codex tools.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a bridge between user prompts and multiple AI generation APIs, creating a surface for indirect prompt injection where untrusted data could influence agent behavior.
Ingestion points: Ingests user-supplied motion and image prompts in creator.py and structured scene data in media.py.
Boundary markers: No explicit delimiters or instructions are used to separate user data from system prompts when calling provider APIs.
Capability inventory: The skill possesses significant capabilities including shell command execution, network access to multiple APIs, and the ability to write/move files on the local file system.
Sanitization: Sanitization is limited to basic media format sniffing (sniff function in creator.py) and path resolution.
[EXTERNAL_DOWNLOADS]: The skill programmatically downloads generated media files (images, video, audio) from various remote provider endpoints.
Evidence: creator.py uses urllib.request to fetch data from URLs provided by Higgsfield, Google, OpenAI, and ElevenLabs APIs.

creator-video