The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill promotes an insecure installation method using a pipe-to-shell pattern (curl -fsSL https://cli.inference.sh | sh). This allows for the execution of remote scripts from a non-trusted source without user inspection or verification.
[COMMAND_EXECUTION]: The skill executes the infsh CLI tool with wildcard arguments (infsh *). This provides a broad surface for command execution on the vendor's platform, including potentially sensitive operations like infsh login for credential management.
[EXTERNAL_DOWNLOADS]: During installation, the skill initiates downloads of binary executables from dist.inference.sh. These downloads originate from a source that is not listed as a well-known or pre-trusted provider.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its processing of untrusted data.
Ingestion points: The code parameter in the JSON input for infsh app run in SKILL.md accepts arbitrary React/TSX code.
Boundary markers: None; the user-provided code is interpolated directly into the shell command payload without delimiters or warnings to the agent to ignore embedded instructions.
Capability inventory: The skill has access to the infsh tool via the shell, which can execute complex rendering logic and interact with the inference.sh API.
Sanitization: There is no evidence of sanitization, validation, or escaping of the React code before it is passed to the execution environment.

remotion-render