code-wiki

为代码仓库增量构建中文 wiki（./wiki/ 目录），梳理架构、数据流，辅助理解和重构。 触发词：理解代码仓库、梳理架构、生成文档/wiki、读懂代码、分析项目结构。 即使未提及"wiki"，只要意图是系统性理解代码仓库或为重构做准备，都应触发。 命令：init, scan, deep, query, lint。

全自动仓库级单元测试生成。扫描源码 → 分析函数 → 生成测试 → 执行并出报告，无需人工干预。 支持全量/增量模式，当前支持 Python（pytest）和 C++（Google Test）。 触发词：单元测试、生成测试、测试覆盖率、增量测试、回归测试、unit test、跑测试、给函数加测试、哪些函数没被测试。 提及 test/generated_unit/ 或 test_cases.json 时也应触发。 命令：init, generate, run, auto。

半自动模块测试/集成测试生成。扫描仓库 → 生成测试配置 → 补充代码引用和测试目标 → 工程师审查 → 生成并运行 pytest/googletest。 触发词：模块测试、集成测试、测试配置、扫描模块、生成测试目标、module test、integration test、test config。 提及 test-config/ 目录或模块测试 YAML 文件时也应触发。 命令：init, generate, run。

Deep analysis skill that pairs a research paper with its open-source implementation to extract core innovations and reveal implementation details the paper doesn't document. Use this whenever a user provides both a paper (PDF/arxiv link) and a code repository, or when they mention wanting to understand "how the model is actually built", "the real architecture", "what the code actually does vs the paper", or want to "reproduce", "reimplement", or "deeply understand" a specific paper's method. Also trigger when the user gives only a paper but asks about implementation details, or only code but asks what the underlying paper says. Keywords that should trigger this: reproduce, reimplement, core module, real architecture, hidden details, paper vs code, ablation, novel contribution, innovation implementation. Do NOT use for generic paper summarization (no code involved) or generic code review (no paper involved).