The Agent Skills Directory

Unverifiable Dependencies & Remote Code Execution (CRITICAL): The skill instructs the agent to execute a command that downloads a script from the internet and pipes it directly into the shell (curl -fsSL https://cli.inference.sh | sh). This is a confirmed remote code execution pattern that allows a third-party server to run arbitrary code on the host system.
External Downloads (HIGH): The domain inference.sh is not included in the list of trusted external sources. Executing software from unverified sources is a high-risk activity that bypasses standard package management and security verification.
Command Execution (HIGH): The skill requests broad permissions to execute the infsh tool with any arguments (Bash(infsh *)). Since this tool is installed via an untrusted RCE pattern, granting the agent these permissions effectively allows it to interact with an unverified and potentially malicious executable.
Indirect Prompt Injection (LOW): The skill exposes a surface for indirect prompt injection via user-provided inputs.
Ingestion points: User prompts are embedded within the JSON input for the infsh app run command (e.g., input.json or inline JSON strings).
Boundary markers: No boundary markers or 'ignore' instructions are present to prevent the agent from being influenced by instructions hidden within the video generation prompts.
Capability inventory: The agent has permission to run the infsh command which communicates with external APIs and executes local logic.
Sanitization: There is no evidence of escaping or validation performed on the prompt before it is passed to the CLI tool.

google-veo