The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The SKILL.md file contains a bootstrap instruction: curl -fsSL https://raw.githubusercontent.com/higgsfield-ai/cli/main/install.sh | sh. This pattern is extremely dangerous as it executes unverified code from the higgsfield-ai GitHub repository directly in the user's shell environment. Upstream compromise of this repository would result in immediate shell execution on the host system.
[COMMAND_EXECUTION]: The skill relies extensively on shell command execution via the Bash tool to process game assets, run local servers, and manage build pipelines. This includes execution of internal Python scripts such as scripts/pipeline.py and complex command chains in the reference documentation.
[EXTERNAL_DOWNLOADS]: The skill downloads software and assets from external domains, including a Blender binary from download.blender.org and character asset packs from third-party GitHub repositories like KayKit-Game-Assets.
[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection.
Ingestion points: Untrusted user input is ingested via game briefs and stored in project files like design/assets.csv.
Boundary markers: There are no explicit instructions or delimiters to isolate or ignore embedded instructions within this user data.
Capability inventory: The skill has broad capabilities including shell access, file system manipulation, and network operations.
Sanitization: There is no evidence of filtering or escaping of the user-provided strings before they are incorporated into prompts for the Higgsfield generation models.

higgsfield-game-generation