The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The scripts/connect.py file (line 70) contains a hardcoded command to download a script from https://raw.githubusercontent.com/higgsfield-ai/cli/main/install.sh and pipe it directly to sh. This allows arbitrary code execution from a remote source that is not listed as a trusted vendor. The command is executed in line 253 when a user confirms installation.
[COMMAND_EXECUTION]: In scripts/connect.py (line 253), the skill uses subprocess.run with shell=True to execute installation strings for Higgsfield and Codex. Using a shell for execution is a security risk as it can facilitate command injection and other side effects.
[INDIRECT_PROMPT_INJECTION]: The scripts/creator.py script interpolates the user-provided --prompt directly into instructions for sub-agents used by the agy (line 504) and codex (line 529) tools. This creates an attack surface where a malicious user prompt could hijack the sub-agent's session.
Ingestion points: The user-provided prompt argument in scripts/creator.py.
Boundary markers: No delimiters or safety instructions are used to separate the user-provided prompt from the agent's internal instructions.
Capability inventory: The skill possesses the ability to run shell commands, interact with the file system, and communicate with external AI providers.
Sanitization: There is no escaping, validation, or filtering performed on the user prompt before it is sent to external agents.

creator-image