v1.0.0

Initial release — v1.0.0

This is the initial public release of the reproduction package for the paper "Large Language Models as a Decision-making Core for Semantic Capability Check in Asset Administration Shell-based Intelligent Manufacturing."

Capability Check module (src/capability_check.py) — implements the core algorithm that receives a provider and a requester AAS capability submodel, preprocesses them (stripping non-English descriptions, qualifiers, semantic IDs, and model type metadata), renders the system and user prompts, and calls the LLM via an OpenAI-compatible API. Returns a structured result with the capable decision, the model's justification, token usage, and the raw output.

Test suite (tests/) — four test groups (G1–G4) covering 454 balanced positive/negative cases for semantic capability matching, insufficient inferential evidence, limiting properties and parameterization, and invalid or inconsistent data. Each group is parametrized from the corresponding dataset file. The conftest.py integrates with pytest-repeat and pytest-xdist, collecting per-test results and writing timestamped JSON reports, a text summary, and a cumulative benchmark history CSV at the end of each session.

Dataset (dataset/) — four JSON files containing the full test case corpus derived from Screwing capability submodel instances, with base instances built in AASX Package Explorer and extended through automated parametric variation and manual structural editing.

Prompts (prompts/) — system and user prompt templates used by the capability check module, with placeholders for the provider and requester submodels.