The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill documentation describes a proxy mechanism that routes API requests to various local command-line interfaces (CLIs) including claude, gemini, qwen, kilo, opencode, and a pi CLI. This pattern involves the dynamic invocation of local system binaries based on model aliases and effort levels provided in the network request.
[COMMAND_EXECUTION]: The skill manages and interacts with local services through a systemd user unit (cli-proxy.service). The routing logic translates API calls into shell commands executed against third-party CLI tools.
[INDIRECT_PROMPT_INJECTION]: The skill exposes a local network endpoint at http://127.0.0.1:7890/v1 which acts as a bridge between untrusted external data and internal system tools. \n
Ingestion points: Network endpoint at 127.0.0.1:7890/v1 accepting chat completion payloads. \n
Boundary markers: The skill explicitly instructs to fold system-role messages into user turns for specific providers, which modifies the structure of the input but provides no security boundaries. \n
Capability inventory: The skill has the capability to execute multiple local binaries (CLIs) and perform external network operations to the MiniMax API. \n
Sanitization: There is no evidence of input validation or sanitization for the messages being passed to the underlying CLI tools.

cli-proxy