hanzhangzzz

harness

Minimal Harness Engineering loop for a code repository: three roles — Inspector finds issues, Worker fixes them, Reviewer verifies — coordinated through a shared TODO.md board, runnable once or on a schedule. Use when the user wants an agent to continuously inspect, fix and review a codebase, or says /harness, harness engineering, 启动 harness, AI 自治循环, 自动巡检修复, 定时改进代码.

hkr-render

WeChat Official Account (微信公众号) publishing pipeline: format Markdown or plain text into WeChat-compatible inline-style HTML with 7 polished themes (dark mode and mobile font sizes tuned), optionally generate a cover image with brightness check and auto-darkening, and push one or many articles to the WeChat draft box (多图文). Use when the user wants to format, typeset or publish a WeChat article, or says 排版, 微信排版, 公众号排版, 公众号发布, 格式化文章, format, /format. Requires a config.json with WeChat AppID/Secret (see SKILL.md).

x-article-download

Download X/Twitter content to local Markdown: a single tweet or long-form article (text, images, videos, linked GitHub repos) or an entire account in batch; auto-detects the content type and applies the right strategy. Use when the user shares an x.com or twitter.com link and wants to save, archive, analyze or translate it, or says 下载推文, 下载 X 文章, 保存这条推特, 批量下载账号, download tweet, archive X account, twitter to markdown.

xiaohongshu-downloader

Download a Xiaohongshu (小红书 / RedNote) video from a xiaohongshu.com or xhslink.com link and transcribe its voice-over with Whisper into a local Markdown transcript (Chinese kept verbatim, English translated to Chinese), saved in a complete folder structure. Use when the user shares a Xiaohongshu video link and wants the spoken content as text for analysis, summary or translation, or says 下载小红书, 小红书视频转文字, 小红书口播稿, 提取口播, xiaohongshu transcript, RedNote video to text.

do-something

Autonomously push the current project forward: answer feedback first, select the highest-value task supported by real evidence, finish and verify a durable outcome, or explicitly return NO-OP when nothing valuable can truly be completed. Every run continues on one do/main branch with bounded DO.md state; optional MR mode pairs with ci-review's separate value and execution verdicts before auto-merge. Built for unattended cron/loop runs and manual triggers. Use when the user says /do-something, 做点什么, 自己看着办, 推进一下, do something useful, keep improving this project.

ci-review

Install a CI-triggered LLM code reviewer into the current repo: on every push to a PR/MR, Claude Code runs headless, reproduces every claim the change makes, hunts correctness bugs with concrete failure scenarios, and posts a machine-readable execution verdict. For configured bot branches (do/* by default), it also requires a separate value verdict before auto-merge, rejecting bookkeeping-only work and incomplete outcomes. Detects GitHub vs GitLab, installs a deterministic verdict gate, and supports review-only or auto-merge tiers. Use when the user says /ci-review, 装一个 CI 代码审查, CI 里自动 code review, 给 PR/MR 加 AI review, 让机器人验证 PR, 审查通过自动合并.

hook-test-kit

为 Claude Code hook 脚本（PreToolUse / SessionStart / UserPromptSubmit 等）生成行为矩阵测试骨架并指导填场景：scratch fixture 隔离、stdin 喂 JSON、EMPTY/!否定/正则断言协议、已知 bash 坑位预修（路径混入子命令、$var 后跟全角字符、case 变量 pattern 不解析交替）。Use when 用户写了或改了 hook 脚本要测试，说测一下这个 hook、hook 测试、hook-test-kit、给 hook 补测试。

gpt-image2-prompt-director

Turn a weak idea into a high-impact GPT image2 (gpt-image-2) generation brief: prompt director workflows, idea generation, visual prompt frameworks, artifact specs for avatars, sticker/emoji packs, infographics, platform covers, posters and product shots, plus benchmark evaluation against an expert prompt set. Use when the user wants to create, improve, repair or systematically evaluate image generation prompts, or says $gpt-image2-prompt-director, 生图提示词, 图片提示词优化, 头像提示词, 表情包, 信息图提示词, 海报提示词, 修一下这个 prompt.

md2view

Re-encode a Markdown document into a traceable human-reading view: a single-file HTML with the original on the left and a model-designed visual re-organization on the right (architecture diagrams, flow chains, comparison matrices, argument maps, annotated cards), where every element carries data-sources anchors back to the source, visible text keeps lexical anchors and numbers are copied verbatim. Use when the user wants a retrospective, report, spec, README or long technical document turned into something easier to absorb and share, or says /md2view, md2view, 信息重组, 双栏阅读, 文档可视化, 把文档做成图. Do not use for plain faithful Markdown rendering.

repo-map

Local repository map for multi-repo development: scan every local git repo into an incremental, self-healing index (name, path, read/write role, purpose); a UserPromptSubmit hook auto-injects the path and role whenever a repo name is mentioned, so cross-repo references and edits never need hand-pasted paths; optional macOS launchd scheduled sync keeps new repos in the map. Use when the user works across multiple local repositories or says 仓库地图, repo-map, repo map, 全局项目索引, 跨仓库引用, 找不到本地仓库, 仓库地图定时更新.

wechat-article-md-local

Download and read a single WeChat Official Account article (微信公众号文章) from a public mp.weixin.qq.com link, save it in the workspace as local Markdown with all images downloaded locally, falling back to HTML when Markdown conversion is impossible. ALWAYS use this skill first whenever the user sends a single WeChat article link — even without asking to download it — and especially before summarizing, extracting, analyzing, quoting, rewriting or translating that article. Triggers: mp.weixin.qq.com link, 下载公众号文章, 公众号文章转 Markdown, 保存这篇公众号, wechat article to markdown.

repo-tidy

Git repository tidy-up and parallel-task base: switch back to the latest master/main, delete merged or upstream-gone branches, remove stale worktrees; `--new <task>` does tidy + create a task branch in one command and auto-creates a parallel worktree when the main checkout is busy; a SessionStart hook injects repo status when a session starts. Use before a new repository-changing task or when the user explicitly requests cleanup. Read-only audits, questions, continued tasks, and [repo-status] alone do not trigger mutations. Also use when the user says 归位, 整理仓库, 清理分支, 清理 worktree, 开新任务, repo tidy, tidy repo, clean branches.

scanned-book-ocr

将纯图片或扫描版书籍 PDF 转为确定性、页码可追溯的文本，并在不降低识别质量的前提下基准测试并发、执行完整性校验和视觉抽检。用于整本扫描书 OCR、OCR 工具选型落地和给后续拆书/检索准备可信文本；不用于已有可靠文字层的普通 PDF，也不把未经回看原页的 OCR 当作精确引文。

doc-reader

Translate an English technical article (web URL) or a local PDF into accurate Chinese, section by section with consistent terminology, keeping 100% of the images, and build a local three-column preview HTML (original · translation · AI slides). Slides are generated through the local Codex CLI built-in imagegen, so no image API key is needed. Use when the user shares an English blog post, paper, docs page or PDF and wants a faithful, side-by-side Chinese version, or says /doc-reader, 翻译这篇文章, 翻译文档, 翻译博客, 翻译 PDF, translate this article. Options: --no-ppt skips slide generation, --quick translates only.