The Agent Skills Directory

REMOTE_CODE_EXECUTION (HIGH): The skill documentation and quick start guide instruct the agent to run curl -fsSL https://cli.inference.sh | sh. This is a critical security risk as it downloads and executes an arbitrary script from a non-whitelisted domain directly in the shell environment without prior inspection or verification.
COMMAND_EXECUTION (MEDIUM): The skill requires the Bash tool with permissions to execute infsh *. This allows the agent to run any sub-command or flag provided by the external CLI, which could include file system access or network operations depending on the CLI's capabilities.
EXTERNAL_DOWNLOADS (MEDIUM): The installation script downloads binaries from dist.inference.sh. Since this domain is not in the trusted source list, the integrity of these downloads cannot be verified by the system, posing a supply-chain risk.
PROMPT_INJECTION (LOW): The skill is vulnerable to Indirect Prompt Injection (Category 8).
Ingestion points: Untrusted user input is interpolated directly into the text field of JSON payloads for infsh app run commands.
Boundary markers: Absent. The skill does not use delimiters (like XML tags or triple backticks) to separate user data from command instructions.
Capability inventory: The skill has Bash access to the infsh utility, which communicates with remote APIs.
Sanitization: Absent. There is no evidence of input validation or escaping before passing data to the shell.

text-to-speech