higgsfield-generate

Fail

Audited by Gen Agent Trust Hub on Oct 5, 2026

Risk Level: HIGHREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill's bootstrap instructions in SKILL.md include a command to download and execute an installation script (curl -fsSL https://raw.githubusercontent.com/higgsfield-ai/cli/main/install.sh | sh) for the Higgsfield CLI. This is the official installation method provided by the service vendor for the tool the skill is designed to manage.
  • [COMMAND_EXECUTION]: The skill uses the Bash tool to perform complex operations, including constructing JSON payloads via printf and mktemp, and executing various higgsfield CLI subcommands for asset generation, product management, and account authentication.
  • [INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection attack surface as it fetches content from user-provided URLs to create 'brand kits' and 'product' profiles. This external data is processed and used to influence downstream generation tasks.
  • Ingestion points: External metadata is ingested via higgsfield marketing-studio brand-kits fetch --url <url> and higgsfield marketing-studio products fetch --url <url>.
  • Boundary markers: The skill does not define explicit delimiters or instructions to ignore embedded prompts within the fetched website content.
  • Capability inventory: The agent has the capability to write files, execute CLI tools (via Bash), and submit processed data to external generation models.
  • Sanitization: The skill instructions do not specify any sanitization or validation steps for the fetched metadata before it is interpolated into generation prompts.
  • [PROMPT_INJECTION]: The skill contains instructions that modify the agent's standard behavior by requiring the suppression of technical narration and internal process updates (e.g., "Don't narrate 'calling higgsfield cost'", "polling job", "No raw IDs"). While intended for a cleaner user experience, this reduces the transparency of the agent's background actions.
  • [EXTERNAL_DOWNLOADS]: The skill fetches configuration and asset data from Higgsfield's official infrastructure and allows importing product information from external landing page URLs.
Recommendations
  • HIGH: Downloads and executes remote code from: https://raw.githubusercontent.com/higgsfield-ai/cli/main/install.sh - DO NOT USE without thorough review
Audit Metadata
Risk Level
HIGH
Analyzed
Oct 5, 2026, 12:39 PM