agent-tools

Pass

Audited by Gen Agent Trust Hub on Apr 27, 2026

Risk Level: SAFEREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill provides instructions to install the belt CLI tool by piping a script from https://cli.inference.sh directly to the shell.
  • [EXTERNAL_DOWNLOADS]: Installation and updates involve downloading binaries and manifests from the vendor's distribution infrastructure at dist.inference.sh.
  • [DATA_EXFILTRATION]: The belt CLI tool facilitates reading local media files and uploading them to the inference.sh cloud service for processing by AI models.
  • [COMMAND_EXECUTION]: The skill relies on the Bash tool to execute belt CLI commands for discovering and running cloud-based AI applications.
  • [PROMPT_INJECTION]: The skill ingests data from external AI model outputs and local user files, creating an indirect prompt injection surface. \n
  • Ingestion points: Resulting JSON and URLs from belt app run commands and local file paths. \n
  • Boundary markers: No protective delimiters are specified to separate external content from agent instructions. \n
  • Capability inventory: Reading local files and network communication via the belt tool. \n
  • Sanitization: No explicit validation or filtering of model-generated content is described.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 27, 2026, 12:42 AM
Security Audit — agent-trust-hub — agent-tools