The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent 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 the shell (| sh). This pattern is a critical security vulnerability as it executes unverified code from an external source without a prior integrity check or review.
[COMMAND_EXECUTION]: The skill extensively uses the Bash tool to interact with the system CLI. It performs high-risk operations including user authentication (higgsfield auth login), local file uploads, and system-level configuration changes during the bootstrap phase.
[EXTERNAL_DOWNLOADS]: The skill performs automated downloads of executable scripts and fetches content from arbitrary user-provided URLs (e.g., product pages or App Store links) to process them via the CLI.
[PROMPT_INJECTION]: The skill interpolates user-supplied text directly into shell command arguments (e.g., higgsfield generate create ... --prompt "[prompt]"). While the instructions wrap the prompt in quotes, they do not specify sanitization or escaping of shell metacharacters (such as backticks, semicolons, or dollar signs), which could allow a malicious user to achieve arbitrary command execution on the host environment.
Ingestion points: User-provided prompts and product URLs defined in SKILL.md.
Boundary markers: The instructions use standard double quotes for argument wrapping but lack explicit warnings to ignore or escape embedded shell instructions.
Capability inventory: The skill has full access to the Bash tool for CLI interactions and file system operations.
Sanitization: No input validation or shell-escaping logic is implemented or requested within the skill's instructions.

higgsfield-generate