xlinsist/academic-skills

reproduce

当用户希望你在一个代码仓库中完成可复现的构建、实验、对比评测和报告沉淀时使用此 skill。它适用于需要搭建本地环境、运行一个或多个 MVP 实验、记录实验设置与结果、解释性能差异，并输出可复现文档与环境脚本的场景。

analysis-module

Use when users need a structured “system understanding + design analysis” of a specific module in an AI repository, covering interface/data, pipeline role, data construction, supervision signals, model structure, loss/training, design motivation, system…

create-task

Create a GitHub issue that assigns a developer a concrete PR-sized task. Use when the user wants an issue/task description with Deliverables, Background, Task Description, and Timeline so an engineer can execute and submit a specific PR.

原文语言：英语

logic-check

检查论文全文逻辑自洽性和严密性，包括整体结构、实验设计和段落逻辑

polish

润色学术论文文件，自动识别 LaTeX/PDF/Word 等格式，检查逻辑严密性、缩写规范、语法错误和学术规范性

read-paper

深度精读学术论文，从标题分析到审稿人视角全面解析论文价值、方法逻辑和研究贡献

weekly-report

用于撰写周报、实习汇报、阶段小结。默认输出简洁中文结构：`既定目标 / 已完成 / 进行中 / 待讨论`。

plot-figure

Use when users need a standardized grouped bar chart template for baseline-vs-method comparisons, with fixed axis/legend/color/data-label conventions and user-provided methods/data.

consumer

从 TASK_QUEUE.md 循环拉取任务并执行；闭环后从队列移除，并将完整过程沉淀到 TASK_RECORD.md。

producer

与用户对话创建可执行任务；在 plan mode 下明确目标、方法、验证手段并控制复杂度，然后将任务追加到 TASK_QUEUE.md。

deep-research

Use this skill when conducting topic-based literature surveys with OpenAlex-first retrieval, structured paper synthesis, and repeatable GUIDE-to-report workflow.