infsh-cli
Fail
Audited by Gen Agent Trust Hub on May 17, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill provides instructions to install the
beltCLI using a shell-piped remote script fromhttps://cli.inference.sh. The skill also details a manual installation method that includes SHA-256 checksum and Sigstore signature verification for enhanced security. - [EXTERNAL_DOWNLOADS]: The CLI installer and manual installation process fetch binary assets and metadata from
dist.inference.sh. - [COMMAND_EXECUTION]: The skill makes extensive use of the
Bashtool to executebeltCLI commands for running models and managing AI tasks. It also suggests writing shell completion files to system directories, which may require elevated permissions. - [DATA_EXFILTRATION]: The CLI tool is designed to upload local files (e.g., images, video, audio) to the inference.sh cloud infrastructure. This is an intended and documented feature required for processing local assets with cloud-based AI models.
- [PROMPT_INJECTION]: An indirect prompt injection surface exists as the agent processes data returned from external AI applications. Ingestion points: Results from
belt app runandbelt task getare incorporated into the agent context. Boundary markers: No specific delimiters or instructions to ignore embedded content are provided. Capability inventory: The agent has access to theBashtool to executebeltcommands. Sanitization: No explicit validation or filtering of the remote application output is described.
Recommendations
- HIGH: Downloads and executes remote code from: https://cli.inference.sh - DO NOT USE without thorough review
Audit Metadata