rivet-agentos
Pass
Audited by Gen Agent Trust Hub on Oct 26, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill is a legitimate collection of documentation and code examples for the agentOS platform. The behavior observed aligns perfectly with the stated purpose of providing a developer reference for building agent environments.
- [EXTERNAL_DOWNLOADS]: The skill demonstrates how to fetch and install software artifacts from the vendor's release infrastructure (rivet.dev). These operations implement security best practices, including manifest-based discovery and cryptographic digest verification (expectedDigest) to ensure integrity before installation.
- [COMMAND_EXECUTION]: The library provides APIs for shell command execution and process management within virtualized environments. These capabilities are mediated by a granular permission system (e.g., childProcess permissions) and resource monitoring as shown in the embedded limits and permissions examples.
- [DYNAMIC_EXECUTION]: The skill facilitates the execution of JavaScript, TypeScript, and Python code. This dynamic execution is the core intended functionality of the platform and is protected by isolation boundaries, including process sidecars and configurable V8 heap limits.
- [INDIRECT_PROMPT_INJECTION]: The skill provides an execution surface for agent-generated code (Ingestion points). The platform implements mandatory isolation boundaries and permission scoping (Boundary markers) to restrict access to the host system. The available capabilities include process execution, filesystem access, and network operations (Capability inventory), which are mediated by a secure architecture using Zod-validated host functions for safe input processing (Sanitization).
Audit Metadata