The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The bootstrap process in SKILL.md instructs the agent to download and execute an installation script (install.sh) from the vendor's official GitHub repository (higgsfield-ai/cli) using a shell pipe (curl | sh) if the CLI is not present on the system.
[COMMAND_EXECUTION]: The skill relies extensively on the Bash tool to execute higgsfield CLI commands for asset generation, product fetching, and account management. It also uses utility commands like mktemp and printf to manage temporary JSON configurations for the CLI.
[EXTERNAL_DOWNLOADS]: The skill fetches content from external product URLs using the higgsfield marketing-studio products fetch command to import metadata for advertising workflows.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its data processing model.
Ingestion points: The skill ingests untrusted data from user-provided prompts and external website content fetched via URLs in SKILL.md and references/marketing-products.md.
Boundary markers: There are no explicit delimiters or "ignore instructions" markers used when interpolating external data into CLI arguments or the agent's context.
Capability inventory: The skill possesses the capability to execute shell commands, perform network operations, and write to the local filesystem via the Bash tool.
Sanitization: No sanitization or validation logic is present to filter potentially malicious instructions embedded in the external content or user prompts before processing.
[PROMPT_INJECTION]: The skill contains highly specific UX instructions in SKILL.md that direct the agent to suppress certain outputs (e.g., 'Don't narrate', 'No raw IDs', 'No internal jargon'). While intended for user experience, these instructions represent a significant override of default agent behavior.

higgsfield-generate