LouisHouse5

以标准模板 docx 为基准，批量修复目标 docx 文件的所有格式（显式格式 + 隐藏格式），确保格式逐节、逐行、逐表完全一致。支持段落、表格、页面、分节、页眉页脚、目录域、样式定义等全维度对齐

Context hand-off when usage is high — summarize the current task, write a hand-off prompt, review it adversarially, and persist it to ~/tmp so a fresh session can resume. Runs a fixed 4-step flow: summarize -> write prompt -> review -> save. Triggered automatically at ~60% context usage (on CLIs whose hook injects a prompt) or manually via this CLI's handoff entry (/handoff, /skill:context-handoff, $context-handoff, etc.). Use when the user says 'handoff', '交接', '上下文交接', '/handoff', or a hook reminds you to call context-handoff.

管理 zcode 平台自身的 MCP 服务器、Skills 发现目录、Memory 存储、Plugins 启用开关——即 zcode 基础设施配置，区别于 GSD 工作流参数（→gsd-config）和 CLAUDE.md/AGENTS.md 内容撰写（→claude-project-config）。当用户说"配置 zcode""加个 MCP server""管理技能""查看/启用插件""看看本项目 memory"".mcp.json 怎么写""zcode.json 在哪""enabledPlugins"等任何针对 zcode 平台配置体系的诉求时触发。注意本技能用 scripts/lib.py 做安全写盘（探测/脱敏/备份/原子替换），第一版对 memory 和 AGENTS.md 只读诊断。

SkillOpt 训练器：把 Agent 技能文档当成可训练状态，通过轨迹反馈、受控文本编辑和验证集门控自动优化 skill，不改模型权重、不增加部署成本

Use this skill for ALL image file processing tasks. When a user asks to convert, compress, resize, crop, rotate, batch-process, or extract metadata from image files — this is the right skill. Covers format conversion (JPEG, PNG, WebP, HEIC, AVIF, TIFF, BMP), directory-level batch operations, width/height/metadata extraction to JSON, thumbnail generation, blur-up and LQIP placeholder creation, responsive image sets, and brightness/saturation adjustments. Works even if the user never mentions Bun. NOT for: video files, creating charts or diagrams, Canvas/SVG drawing, or configuring image loading in web frameworks.