cloudbase-agent-python

Pass

Audited by Gen Agent Trust Hub on Aug 29, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is specifically designed to create agents that ingest and process untrusted user data, creating a surface for indirect prompt injection attacks.
  • Ingestion points: Untrusted data enters the agent context via the messages and state fields in the RunAgentInput model, which is the primary interface for all SDK adapters (e.g., adapter-coze.md, adapter-langgraph.md).
  • Boundary markers: The provided documentation and code examples do not demonstrate the use of delimiters or explicit 'ignore instructions' warnings when interpolating user messages into LLM prompts.
  • Capability inventory: The SDK provides access to high-privilege tools through the cloudbase-agent-tools package, including bash execution (create_bash_tool), filesystem operations (create_read_tool, create_write_tool), and MCP tool integration (referenced in skill.md and references/tools.md).
  • Sanitization: There are no examples or requirements for sanitizing user-provided text before it is passed to the agent's internal logic or tool-calling loops.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 29, 2026, 06:57 AM
Security Audit — agent-trust-hub — cloudbase-agent-python