qutip

Warn

Audited by Gen Agent Trust Hub on Mar 31, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTION
Full Analysis
  • [METADATA_POISONING]: There is a discrepancy between the reported author ('LeonChaoX') and the 'skill-author' field in SKILL.md ('K-Dense Inc.'). Such inconsistencies can be used to mislead users or auditors regarding the origin and trustworthiness of the skill.
  • [UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill instructs users to install qutip, qutip-qip, and qutip-qtrl via the uv pip command. While these are established scientific libraries, the instructions do not specify versions, which could lead to the installation of compromised or malicious versions in a supply chain attack.
  • [DYNAMIC_EXECUTION]: In references/advanced.md, the skill documents the use of time-dependent Hamiltonians provided as strings (e.g., 'cos(w*t)'). QuTiP uses Cython to compile these strings into executable code at runtime to achieve high performance. This mechanism allows for dynamic code generation and execution based on string input.
  • [DYNAMIC_EXECUTION]: The skill provides instructions for saving and loading simulation data using qload and Result.save in references/advanced.md. These functions utilize Python's pickle module, which is vulnerable to arbitrary code execution during deserialization. Loading a malicious .qu or .dat file via these functions could lead to full system compromise.
  • [INDIRECT_PROMPT_INJECTION]: The skill contains multiple surfaces where untrusted data could influence agent behavior or system state.
  • Ingestion points: Input Hamiltonians as strings in sesolve/mesolve and deserialization of saved data files via qload (documented in references/advanced.md).
  • Boundary markers: Absent; the skill does not define delimiters or provide warnings to the agent regarding the separation of user data from executable physics logic.
  • Capability inventory: The skill leverages file system access for saving/loading data and utilizes runtime compilation tools (Cython) for processing dynamics.
  • Sanitization: No sanitization or validation logic is provided for the mathematical strings or serialized files processed by the solvers.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 31, 2026, 08:22 AM
Security Audit — agent-trust-hub — qutip