The Agent Skills Directory

[REMOTE_CODE_EXECUTION] (CRITICAL): The skill instructions include curl -fsSL https://cli.inference.sh | sh. This is a piped remote execution pattern where a script from an untrusted domain is executed directly by the shell, allowing for arbitrary code execution on the host system.
[EXTERNAL_DOWNLOADS] (HIGH): The skill downloads and installs additional components from inference.sh and uses npx to fetch skills from inference-sh/skills, neither of which are within the trusted repository or organization whitelist.
[COMMAND_EXECUTION] (MEDIUM): The skill relies on the infsh command-line tool to perform actions. Since this tool is installed via an unverified remote script, any command executed via infsh inherits the risk of the installation method.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill processes data from external files (like image.json or video.json) and interpolates it into shell commands without sanitization.
Ingestion points: Content from image.json and video.json is used as input for subsequent commands.
Boundary markers: None are present to distinguish between data and instructions.
Capability inventory: The skill has Bash access to execute the infsh tool.
Sanitization: No evidence of input validation or escaping before interpolation into the command string.

twitter-automation