cudaq-guide

Pass

Audited by Gen Agent Trust Hub on Jul 22, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONNO_CODE
Full Analysis
  • [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it retrieves content from local CUDA-Q documentation files using the Read, Glob, and Grep tools.
  • Ingestion points: Local documentation paths (e.g., references/onboarding.md, references/authoring.md) accessed during query resolution.
  • Boundary markers: None identified in the instructions for processing retrieved file content.
  • Capability inventory: The skill provides instructions for shell command execution (installation) and Python kernel authoring.
  • Sanitization: No explicit sanitization or filtering of the retrieved content is documented.
  • [COMMAND_EXECUTION]: The authoring reference (references/authoring.md) includes code patterns for dynamic kernel generation using Python's compile() and exec() functions. While intended for advanced framework use cases, these patterns introduce a risk of arbitrary code execution if user-provided parameters used to construct the dynamic source code are not strictly validated.
  • [NO_CODE]: The skill itself does not contain any executable scripts (Python or Shell) that are run by the agent; it functions primarily as a structured guidance and documentation retrieval system.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 22, 2026, 08:01 AM
Security Audit — agent-trust-hub — cudaq-guide