numpy-low-level

Warn

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: MEDIUMPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • Indirect Prompt Injection (MEDIUM): The skill provides patterns for processing external binary data and files using memory-unsafe functions. If an agent uses these patterns on data from an untrusted source, it creates a vulnerability surface where malformed input could lead to memory-based attacks or system instability.\n
  • Ingestion points: np.frombuffer and np.memmap used to load external content into memory in SKILL.md.\n
  • Boundary markers: No explicit delimiters or instructions to ignore embedded content are present in the provided snippets.\n
  • Capability inventory: Use of as_strided and ctypes.POINTER allows for direct memory access and potential corruption.\n
  • Sanitization: No input validation, shape checking, or stride verification is shown in the examples.\n- Dynamic Execution (MEDIUM): The use of numpy.lib.stride_tricks.as_strided and ctypes allows an agent to bypass standard Python safety guards and directly manipulate raw memory.\n
  • Evidence: The skill encourages the use of as_strided for zero-copy views, which the documentation itself warns can lead to memory corruption or crashes if bounds are miscalculated.\n
  • Evidence: The use of ctypes.data_as exposes raw memory addresses for direct buffer manipulation.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 16, 2026, 12:59 PM