jiangkaiqi2005
- 12 skills
- 0 followers
- 7 hours ago last updated
- ▌ Study Tutor · jiangkaiqi2005 bundleStudy Tutor — a science-based learning assistant for diagnosis, guided teaching, practice, review, spaced repetition, and concise study notes under memory/{subject}-study.md.
- ▌ Guided Learning · jiangkaiqi2005 bundleGuided learning skill — one concept at a time, gentle guidance, comprehension checks, chapter review
- ▌ Vmamba Env Doctor · jiangkaiqi2005 bundleGuide for setting up a stable VMamba environment and resolving common CUDA, PyTorch, MMCV, and compilation errors. Use when users encounter issues like "unsupported GNU version", "libc10.so not found", "No module named 'torch' during pip install", or NumPy/OpenCV version conflicts.
- ▌ Forge Cumcm Solution · jiangkaiqi2005 bundle以三阶段高质量执行完成全国大学生数学建模竞赛（CUMCM、国赛）的审题建模、代码求解与验证、论文写作和提交审查。用于从赛题与附件端到端作答，续作或审计已有模型、代码、结果和论文，检查模型—算法匹配、约束可行性、数据泄漏、结果可复现性、数字单位一致性、图表证据和当年官方模板适配。涉及 MCM/ICM 或一般建模时，仅在用户明确要求沿用本流程时使用。
- ▌ Hcod Result Reporting · jiangkaiqi2005 bundleUse when HyperCOD/VMamba experiment results are ready and Codex needs to archive train logs, choose or confirm an output root, generate train-log HTML visualizations, identify the best iteration, or collect paper-style metrics.
- ▌ Analyze Images With Glm · jiangkaiqi2005 bundleAnalyze local images, screenshots, or image URLs with the free GLM-4.6V-Flash vision model through the BigModel HTTP API, without MCP or third-party Python packages. Use when the active model cannot see images and a task requires visual description, OCR, chart or diagram reading, UI inspection, object identification, image comparison, or other pixel-dependent reasoning.
- ▌ Analyze Images With Qwen Ocr · jiangkaiqi2005 bundleExtract and analyze text, tables, forms, formulas, screenshots, and document images with Qwen3.5-OCR through the DashScope OpenAI-compatible Responses API, without MCP or third-party Python packages. Use when the active model cannot inspect image pixels and the task depends on OCR, structured field extraction, document reading, chart labels, UI text, handwriting, or comparing textual content across images.
- ▌ Analyze Images With Qwen Flash · jiangkaiqi2005 bundleAnalyze local images, screenshots, diagrams, charts, interfaces, photos, and image URLs with the multimodal Qwen3.5-Flash model through the DashScope OpenAI-compatible API, without MCP or third-party Python packages. Use when the active model cannot inspect image pixels and the task requires fast visual description, UI inspection, chart or diagram reasoning, object identification, OCR-assisted understanding, or comparison across images.
- ▌ Structuring Research Workflows · jiangkaiqi2005 bundleUse when a computer vision research method, paper, codebase, model, dataset pipeline, experiment, or deployment system needs to be explained as a rigorous step-by-step workflow or prepared for a publication-style architecture or process diagram.
- ▌ Forge Cumcm Paper · jiangkaiqi2005执行 forge-cumcm-solution 的阶段三：只依据已验证结果和当年官方模板撰写、修改或提交级审查 CUMCM 论文，完成摘要、公式、符号、图表、引用、附件、数字单位一致性、匿名与 AI 披露检查，并由阶段三八席独立专家迭代评审。
- ▌ Forge Cumcm Solving · jiangkaiqi2005执行 forge-cumcm-solution 的阶段二：把已冻结的 CUMCM 模型实现为真实可运行、可复现的代码和求解流程，生成逐问答案、约束证书、独立验证、适用的敏感性与稳健性分析、图表数据和规定结果文件。用于实现、调试、数值审计、统计或机器学习防泄漏和结果证据链检查。
- ▌ Forge Cumcm Modeling · jiangkaiqi2005执行 forge-cumcm-solution 的阶段一：读取 CUMCM 赛题、附件、官方规则和已有材料，完成审题、对象口径、小问关系、核心机制、假设、基础与改进模型、模型接口、方法路由和风险审计，并产出人可读的建模总结。用于新题建模、已有方案复核，或后续阶段发现机制和口径错误后的回退。