CoReasoning: A Competency Model for Collaborative Reasoning with Generative AI

The CoReasoning project, developed by ApartsinProjects, was released on GitHub on June 1, 2026 (under the MIT open-source license). It proposes an assessable human-AI collaborative reasoning competency model for the generative AI era, decomposing it into three dimensions: framework construction, critical judgment, and guidance & optimization. An accompanying open-source online learning platform, CoReasoning Lab, has been developed to provide a theoretical framework and practical tools for AI literacy education.

Most current AI education products focus on quickly obtaining answers from AI rather than human-AI collaborative thinking. The traditional view treats AI as a tool: users提出需求, AI provides answers. CoReasoning argues that complex problems require dynamic collaboration: humans define problems, evaluate outputs, and guide optimization, while AI generates candidate solutions. This model requires humans to possess a unique combination of competencies.

The project decomposes collaborative reasoning competencies into three dimensions:

1. Framework Construction: Transform vague problems into precise tasks, including clarifying boundary constraints, identifying assumptions, resolving ambiguities, and decomposing subtasks;
2. Critical Judgment: Examine flaws in AI outputs, including fact-checking, logical evaluation, bias identification, and risk assessment;
3. Guidance & Optimization: Improve outputs through iterative feedback, including identifying gaps, proposing specific suggestions, maintaining goal alignment, and balancing exploration and convergence.

The research team developed an assessment tool with 16 prompts and tested 40 participants. The results showed that the correlation between skills within the same dimension was much higher than across dimensions (approximately 60:1), indicating that the three dimensions are relatively independent and need to be cultivated separately, as they cannot replace each other. This suggests that AI literacy training should cover all dimensions, not just prompt engineering.

Platform architecture: Frontend static pages (HTML/CSS/JS), backend Node.js + Express, database SQLite/PostgreSQL, AI integration with OpenAI/Groq API, Docker deployment; Learning flow: Framework phase (problem transformation) → AI generates initial solution → Judgment + guidance loop → Independent scoring for three dimensions; Feature functions: Challenge mode (practice/assessment), multilingual support, role permissions (student/teacher/admin), course management, PDF reports.

The project uses the MIT open-source license. Open resources include: complete academic papers (HTML + KaTeX), 16 prompt assessment question banks, experimental datasets and analysis scripts, and manually annotated validation studies, supporting other researchers to replicate, verify, and extend the results.

1. Competencies Over Tools: Cultivate meta-competencies rather than specific tool usage to enhance transferability;
2. Critical Thinking Is Indispensable: Resist AI hallucinations, flattery, and other flaws;
3. Iterative Dialogue Is Better Than Single Queries: Cultivate patience and skills for multi-round collaboration;
4. Multi-Dimensional Assessment: Replace simple accuracy measurement with the three-dimensional framework.

The CoReasoning project explores the cultivation of AI collaborative reasoning competencies from both theoretical (three-dimensional model and independence validation) and practical (learning platform and tools) aspects. It provides teaching resources for educators, open data for researchers, and improvement pathways for learners. In the future, human-AI collaboration will become a core skill, and this project provides a solid starting point for competency cultivation in the AI era.