The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the user to install the core CLI tool by piping a remote script directly to the shell (curl -fsSL https://cli.inference.sh | sh). This pattern executes unverified code from the internet, which is a significant security risk.
[COMMAND_EXECUTION]: The skill is designed around the execution of the belt CLI tool. Through the allowed-tools: Bash(belt *) configuration, the agent is granted permission to run various shell commands to interact with the platform.
[DATA_EXFILTRATION]: The skill documents a 'Local File Uploads' feature where the belt CLI tool automatically reads and uploads local files to the vendor's cloud infrastructure when a path (e.g., /path/to/photo.jpg) is provided instead of a URL. This capability allows the agent to send local data to external servers.
[INDIRECT_PROMPT_INJECTION]: The skill processes both local files and outputs from external AI models, creating a vulnerability surface for indirect instructions to influence agent behavior.
Ingestion points: The agent reads local files for upload and processes JSON results from remote AI applications (SKILL.md, references/running-apps.md).
Boundary markers: There are no explicit delimiters or instructions provided to the agent to treat model outputs or file contents as untrusted data.
Capability inventory: The skill combines shell command execution, local file read/upload access, and network communication.
Sanitization: The instructions do not mention any sanitization or validation of external content before it is processed by the agent.
[EXTERNAL_DOWNLOADS]: The skill depends on binaries and metadata fetched from cli.inference.sh and dist.inference.sh. Manual installation steps involving curl and tar are provided as an alternative, but the primary recommendation is the automated script execution.

infsh-cli