taide-botrun

The Chinese GSAT Solver is a specialized agent skill designed to assist with high school level Chinese proficiency examinations, specifically the General Scholastic Ability Test (GSAT). By implementing a layered architecture comprising Question Identification, Knowledge Retrieval, and Reasoning Templates, the tool provides high-accuracy, verifiable answers for complex linguistic and literary tasks. Instead of rote memorization, the agent utilizes dynamic context assembly to deliver explanations that cite authoritative sources like the Ministry of Education's dictionary and classic literary references. It is ideal for students, educators, and developers building automated exam-prep tools.

• Advanced Context Engineering: Uses a structured retrieval process to inject only the most relevant, domain-specific knowledge into the LLM prompt, ensuring high precision in token-constrained environments.

• Multi-Layered Architecture: Separates logic into distinct tiers: Pattern Recognition (using JSONL descriptors), Knowledge Indexing, and Logic Templates (Drools-style reasoning).

• Explainable AI: Every solution includes a step-by-step reasoning process, clear citation of source materials, and logical排除 (exclusion) arguments for incorrect options.

• Wide Domain Coverage: Expert-level handling of character recognition, idiom correction, literary history, authorial analysis, and rhetorical device identification.

• Input/Output: Accepts structured JSON objects containing the question and options; returns a comprehensive JSONL response including the correct answer, analysis of each option, applied rules, and reasoning steps.

• Practical Usage: Optimized for node.js environments via the solver.mjs script, suitable for CLI tools or backend API integration.

• Constraints: Knowledge base is currently optimized for Taiwan's 108 Curriculum guidelines and standard high school textbooks. Performance depends on high-quality vector retrieval or JSONL indexing.

• Dependencies: Requires Node.js and basic LLM interaction wrappers; knowledge files are strictly curated based on academic sources like the MOE Dictionary.