Coding & Development

Turn the current conversation into a spec and publish it to the project issue tracker: no interview, just synthesis of what you've already discussed.

Upgrade one dependency at a time by reading its changelog, planning for breaking changes, updating code and proving the project still builds and passes. Use when asked to bump, update or migrate a package or framework.

Audit an AGENTS.md, CLAUDE.md, Cursor rule or Copilot instructions file for stale commands, broken paths, vague rules, secrets and contradictions. Use when asked to review, trim or check agent instructions.

Checklist for shipping an HTML5 or WebGL browser game and finding its first players. Covers load speed, touch controls, title and Open Graph share tags, free hosting (GitHub Pages, Netlify, Vercel, itch.io) and where to list it (itch.io, Newgrounds, Foundplay, portals). Use when the user wants to publish, deploy, launch or get players for a browser game.

Test-driven development. Use when the user wants to build features or fix bugs test-first, mentions "red-green-refactor", or wants integration tests.

Compact the current conversation into a handoff document for another agent to pick up.

>- Review, test, and score an AI agent skill from its user's perspective using a two independent review passes: an instruction auditor and a hands-on skill tester who evaluates real outputs. The main orchestrator compares evidence, resolves disagreements, and produces the final verdict. Detect critical flaws and recommend prioritized improvements to usefulness, reliability, completeness, and purpose alignment. Use to review SKILL.md files, installed skills, or skill repositories, including $han-skills-review "$learn".

Conduct a retrospective on a coding session.

Grill the user relentlessly about a plan, decision, or idea. Use when the user wants to stress-test their thinking, or uses any 'grill' trigger phrases.

Teach a codebase or feature with source-grounded Markdown lessons in a project .learn folder. Use when the user asks for a codebase learning guide or invokes $han-learn; save a separate lesson for each $han-learn command.

安全分析并整理 Windows 或 macOS 本地文件夹、移动硬盘或历史备份目录，恢复清晰易懂、长期可维护的目录结构。当用户提到整理磁盘/文件夹/下载目录、检测或清理重复文件、归档旧电脑资料、目录太乱想收拾时使用——即使用户没有明确说"整理"。整理前先全量遍历目录树，为每个目录建立语义画像并归纳主题地图，结构从内容本身归纳得出，与按规则/类型/日期套模板的整理工具划清界限。重复文件优先用本机已有的 Czkawka / fclones CLI 检测，不可用自动降级为平台脚本（Windows: PowerShell + SHA256；macOS/Linux: bash + shasum）。默认不删除任何文件，所有修改先生成计划、全程可回滚。

Turn one or more lesson-plan files (PDF, DOCX, text) into a 60–90 minute classroom slide deck that hits every learning objective, with a hands-on web game after each learning goal. Use when a teacher uploads lesson plans and asks for slides, a slideshow, a deck, or a presentation to teach from. Not for after-class parent updates (use report-generator).

Valida out/<caso>.json da auditoria frontsec (formato exato e coerência de pontos, esforço e prioridade) antes de o agente parar.

Code is read far more often than it is written. Maintainability is what separates a 3-year-old codebase that's a joy from one that's a tar pit. Optimize for the next developer, who may be you in six months.

>- SOLID CUDA 13 / GPU pipeline guardrails for Python+CUDA apps (e.g. NVDEC): concrete typing at boundaries, small modules, explicit device/resource lifetimes, no silent Any leakage. Use when editing CUDA 13, PyNvCodec, GPU decode/inference code.

Our backend services follow **Package by Feature** organized across four sibling modules. This document is the source of truth for where new code goes.

>- SOLID Batch (.bat/.cmd) guardrails: thin wrappers around PowerShell/Python, setlocal, explicit errorlevel checks, no secrets. Use when editing Windows launchers or portable package entrypoints.

本项目的编程学习、分步实作、工程设计与追问复习。用于教我、带我学、复习、grill me、阶段面试及学习中的继续、改好了、答题和提示；也用于维护 learning-coach。普通交付不强制辅导。

适用于中文写作、文档润色、改写、基于材料成稿，以及改善信息组织、论证和语言表达的请求。围绕核心问题与作者想法，在同一份工作文档上分阶段选择、组织和修改信息，按需补充依据，读取已保存正文逐句校验逻辑并定向回退，保留原意与作者风格。用户要求去除“AI味”、讲清重点或让文章更容易理解时使用。

参考文献的 GB/T 7714-2015 著录规范化，以及 URL / DOI / PMID / PMC 链接的 可访问性批量核验与"假失效"识别。 触发场景： - 论文 / 综述 / 作业要交，需要核对参考文献格式 - "帮我按国标改一下参考文献" - 要交付一份文献清单，需要确认链接没死 - 已经有一堆 links，要批量验证哪些还能打开 含一个可复用脚本 `scripts/check_refs.py`，能自动判别 URL / DOI / PMID / PMC 并并发核验，且能区分「真死」与「被反爬」——后者不该判为失效。 不适用：从零生成一份带引文的正文（那是写作任务）； Zotero 样式定制（见正文第六节，按需手工改 CSL）。

按学术论文规范（小三黑体一级标题/四号黑体二级/小四正文宋体1.5倍行距/三线表/中西文混排Times New Roman/统计量斜体） 批量重整 Markdown 或已有 docx 的格式。用户说"按学术格式排版""三线表""小三号黑体标题""1.5倍行距" "数字英文用Times New Roman""P值斜体"时使用。基于 python-docx 直接生成 docx，不走逐段插入。 不适用：高校课程论文/作业的字体字号约定（标准不同）；党政机关公文国标 GB/T 9704 （又是另一套标准）。三套标准不可混用。

从商品参考视频、截图或素材制作可运行的三维或序列帧展示页，支持滚动叙事、鼠标与触屏拖拽旋转、结构展开及小程序风格 H5 界面，提供可编辑模板与浏览器验证。适用于新增商品展示或保留现有交互的增量修改。

>- Pre-commit and pre-merge code and documentation review methodology for Diosix. Activate this skill before committing, merging, or signing off on any feature, refactoring, bug fix, or documentation update, or when the user asks for a code/security/architectural review.

Resume work from a handoff document. Loads the latest handoff in the current repo's handoffs/ directory, or a specific handoff if a path is passed.