The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill contains a 'Bootstrap' step in SKILL.md that instructs the agent to download and execute a script from https://raw.githubusercontent.com/higgsfield-ai/cli/main/install.sh by piping it directly to the shell (| sh). This is a critical security risk as the remote script could be modified to perform malicious actions on the user's system without any validation or review.
[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to interact with the Higgsfield CLI. It performs operations such as creating temporary files (mktemp), writing data to those files, and executing complex CLI commands with multiple flags. This level of system access increases the impact of any potential compromise.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to how it handles external data.
Ingestion points: User-provided text prompts are passed directly to the --prompt flag of the CLI. Additionally, product URLs are fetched using higgsfield marketing-studio products fetch --url, which involves the agent processing content from an attacker-controlled external website.
Boundary markers: There are no explicit boundary markers or instructions to ignore embedded commands within the ingested prompts or URL data.
Capability inventory: The agent has full access to the Bash tool, allowing for file system modifications, network requests, and command execution via the Higgsfield CLI.
Sanitization: The instructions do not specify any sanitization, escaping, or validation of the user input or fetched content before it is used in shell commands.

higgsfield-generate