cloudbase-agent
Warn
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The SDK documentation describes built-in support for tools that execute shell commands (
create_bash_tool) and arbitrary code. These capabilities allow an agent to interact with the underlying system, which represents a significant risk factor if the agent's input is not properly sanitized or if the agent lacks restricted permissions. - [INDIRECT_PROMPT_INJECTION]: The architecture of building agents that process user messages to trigger tool calls creates a surface for indirect prompt injection. The skill documentation addresses this by providing patterns for human-in-the-loop verification (using
interrupt()) and strictly managed state contexts. - [DYNAMIC_EXECUTION]: The skill facilitates the dynamic creation and execution of tool logic based on AI model outputs, including the ability to route and execute code chunks and system-level operations within the agent's reasoning loop.
- [EXTERNAL_DOWNLOADS]: Deployment instructions guide users to download and install dependencies from public package registries (npm and PyPI). The referenced packages include SDK components (@cloudbase/agent-*) and well-known AI libraries (LangGraph, LangChain, OpenAI), which are standard resources for this development context.
Audit Metadata