The Agent Skills Directory

Remote Code Execution (HIGH): The skill instructs the execution of curl -fsSL https://cli.inference.sh | sh. This is a critical security risk as it pipes a remote script directly into a shell interpreter without prior verification. An attacker who compromises the source domain could achieve full system compromise.
External Downloads (MEDIUM): The skill relies on an external binary (infsh) downloaded from dist.inference.sh. This source is not included in the list of trusted external organizations, and the binary exists outside of standard, audited package registries like PyPI or NPM.
Command Execution (MEDIUM): The skill is granted broad access to the infsh command via Bash(infsh *). While scoped to a specific binary, the lack of input sanitization and the nature of the tool (which interacts with remote APIs and manages logins) increases the attack surface.
Indirect Prompt Injection (LOW):
Ingestion points: User-provided prompt and image_url values are interpolated into infsh command arguments.
Boundary markers: Absent. The skill uses raw JSON strings within shell commands to pass user input.
Capability inventory: The skill can execute the infsh tool, which performs network operations and likely handles authentication tokens (infsh login).
Sanitization: No sanitization or escaping of the user-provided prompt is performed before it is passed to the shell command, allowing for potential command argument injection if the agent is not careful.

flux-image