The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions guide the user to install the qutip library (v5.3.0) and its extension packages (qutip-qip, qutip-qtrl, qutip-jax) using the uv package manager. These are well-established scientific computing packages hosted on official registries.
[COMMAND_EXECUTION]: The skill includes several utility scripts (e.g., qobj_model_validator.py, two_level_simulation.py) and provides specific command-line instructions for their execution to perform quantum simulations and audit results locally.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data via JSON model and result files, which represents a potential injection surface.
Ingestion points: Data is loaded through the load_json_object function in scripts/_common.py, which is utilized by scripts such as qobj_model_validator.py and result_audit.py to ingest user-supplied model and result files.
Boundary markers: The skill explicitly instructs agents to record model assumptions and record units/conventions before solving. However, there are no specific prompt delimiters mentioned for the ingested file content itself.
Capability inventory: The skill possesses the capability to perform local numerical simulations and write JSON results to the local filesystem. It does not perform network operations or execute arbitrary shell commands based on the ingested data.
Sanitization: The _common.py utility implements robust sanitization by using a restricted JSON parser that rejects non-standard constants and duplicate keys. Furthermore, checked_input_file enforces strict bounds, including a 1MB size limit and the rejection of symlinks and network URLs.
[SAFE]: The skill follows security best practices by avoiding unsafe serialization (like pickle), explicitly disclaiming dynamic code execution, and implementing strict, bounded I/O for all local script operations.

qutip