The Agent Skills Directory

[SAFE]: The skill implements rigorous security boundaries for local file operations. The scripts/_common.py utility includes functions like checked_input_file and checked_output_file that explicitly reject network-based URLs and symbolic links, effectively preventing unauthorized data access or exfiltration.
[SAFE]: Data ingestion is heavily sanitized and validated. The load_json_object function uses a strict JSON parser that rejects non-standard constants (such as NaN or Infinity) and duplicate keys. Furthermore, scripts/qobj_model_validator.py performs a structural preflight audit to ensure that only numeric matrices and vectors are processed, preventing injection through malformed data fields.
[SAFE]: The skill explicitly mitigates risks associated with dynamic execution and unsafe deserialization. Documentation and scripts (e.g., references/time_evolution.md) emphasize the use of Python callables or numeric arrays rather than QuTiP's string-based coefficient parsing. Additionally, scripts/result_audit.py audits simulation results without using pickle or loading Python objects, protecting against RCE via untrusted data.
[SAFE]: Dependency management is transparent and limited to well-known scientific distributions. The skill specifies pinned versions for qutip, numpy, and scipy, and it provides explicit warnings against using unreleased or experimental repositories (such as qutip-cupy), demonstrating high supply chain awareness.

qutip