The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill's bootstrap process executes a remote script from GitHub by piping it directly to the shell: curl -fsSL https://raw.githubusercontent.com/higgsfield-ai/cli/main/install.sh | sh. This pattern is inherently dangerous as it executes unverified code from an external source.
[PROMPT_INJECTION]: The skill is vulnerable to command injection because it instructs the agent to place user-provided interview responses directly into the --prompt argument of a Bash command. A malicious user could provide shell metacharacters (e.g., ;, &&, or backticks) to execute unauthorized local commands.
Ingestion points: User descriptions from the 'Pre-generation interview' in SKILL.md are used as CLI arguments.
Boundary markers: None. The instructions do not provide delimiters or instructions for the agent to treat the user input as non-executable data.
Capability inventory: The skill explicitly allows Bash via allowed-tools, providing a broad attack surface for command execution.
Sanitization: There are no instructions to sanitize, escape, or validate the user-provided strings before they are interpolated into the shell command.
[COMMAND_EXECUTION]: The skill frequently uses the Bash tool to run CLI commands for image generation (higgsfield product-photoshoot create) and authentication (higgsfield auth login).
[EXTERNAL_DOWNLOADS]: The skill downloads resources from the higgsfield-ai GitHub organization during the bootstrap phase.

higgsfield-product-photoshoot