The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill's primary function is to send and execute arbitrary Python code to the inference.sh cloud environment via the belt CLI tool.
[COMMAND_EXECUTION]: The skill utilizes the belt command-line interface to interact with the remote service, requiring the agent to execute shell commands with user-influenced arguments.
[EXTERNAL_DOWNLOADS]: The skill references and encourages the download of installation scripts from github.com/inference-sh, which is an external source not included in the primary trusted list.
[DATA_EXFILTRATION]: The remote execution environment is pre-installed with network-capable libraries such as requests, httpx, selenium, and playwright, which allow the code to transmit data to arbitrary external endpoints.
[PROMPT_INJECTION]: The skill provides a surface for indirect prompt injection (Category 8) where an attacker could influence the agent to generate and execute malicious Python scripts.
Ingestion points: The code field in the JSON input for the belt app run command.
Boundary markers: None found. There are no instructions or delimiters provided to ensure the agent handles user-influenced code safely.
Capability inventory: The execution environment has full Python capabilities, including network access (via requests), browser automation (via playwright), and file system access within the sandbox.
Sanitization: No sanitization or validation mechanisms are described to inspect the code before execution.

python-executor