MLliu6
UserGeneral, project-agnostic skills for Claude Code · 面向 Claude Code 的通用技能集合(中英双语)
Categories
Indexed Skills (8)
claude-account-sync
Mirror a workspace AND all Claude Code conversation history (--resume sessions), path references, and resume ordering from one user account to another on the same SSH host, so `claude --resume` on the destination picks up exactly where the source left off. Use this whenever you switch between two accounts on a shared host — because one ran out of quota/credits, or when the user says "sync my account", "switch accounts", "move my conversation history over", or "my resume history is missing after switching". Direction is a parameter (--from / --to), so reach for it in EITHER direction; it also restores a workspace onto a fresh account.
claude-session-doctor
Diagnose and fix common Claude Code session and disk problems. Reach for it when the disk is full or a command dies with ENOSPC / "no space left on device" / "/tmp is full"; when `claude --resume` history is missing, reverted, empty, or shows sessions in the wrong order (an old chat on top, the latest one buried); or when a session transcript looks truncated or corrupt. Trigger phrases: "ENOSPC", "no space left on device", "disk full", "/tmp full", "resume history missing", "resume order wrong", "session corrupt/truncated", 以及 "磁盘满了", "空间不足", "resume 历史丢了", "会话顺序不对", "记录损坏/截断". Everything is READ-ONLY except one explicit --apply that only rewrites file modification times.
fix-ssh-terminal
Diagnose and fix the "an idle SSH session dies and afterwards the terminal spews escape codes / mouse garbage" failure. Symptoms: you leave a full-screen TUI (an AI coding CLI, vim, htop, a REPL, ...) idle over SSH; when you return and touch it you get "connection reset by peer" and are dropped back to the machine you SSH'd from; afterwards moving the mouse or clicking prints garbage like [<0;80;24M (staying still is fine, only movement triggers it), and you must run reset and log in again. Root cause: the idle SSH connection is torn down by the network while a TUI holds the terminal in raw + mouse-reporting mode, and it is not running inside tmux/screen, so it is killed before it can restore the terminal. Trigger this skill when the user mentions SSH disconnect, connection reset by peer, dropped/idle SSH, needing reset after being idle, mouse codes / escape codes / garbage on screen, or wanting a long-running TUI to survive disconnects. Chinese triggers: SSH 掉线, peer reset, 连接被重置, 终端乱码, 光标移动出现字母数字码, 空闲掉线, 希望
goal-mode-loop
Run a long-running Claude Code conversation as an autonomous "GOAL MODE" loop: pin down one goal with explicit, checkable success criteria; break the work into stages; verify each stage with evidence before advancing; checkpoint progress to a file so it survives context loss; and know when to stop or hand off. Reach for it when the user says "work in goal mode", "keep iterating until it's done", "run this autonomously", "set up a progress checkpoint", "don't stop until the tests pass", or (中文) "进入目标模式", "自动持续迭代", "分阶段推进并逐阶段验证", "把进度记录到文件", "一直做到达成目标". Complements the `handoff` skill: goal-mode-loop operates the running chat; handoff writes the transfer document when you pause.
handoff
Produce an exhaustive HANDOFF DOCUMENT plus a ready-to-paste CONTINUATION PROMPT so a brand-new Claude Code conversation can resume long-running, goal-driven development without losing context — and keep working at production / open-source quality toward a mergeable pull request. Reach for this whenever the current chat is getting long, context is filling up, or the work must move to a fresh session or another person. Trigger phrases: "hand off this work", "write a handoff", "handoff doc", "continuation prompt", "resume this in a new conversation", "context is almost full, summarize state", "pick up where I left off", "写个交接文档", "生成接续提示词", "把当前进度交接给新会话", "上下文快满了,先把状态存下来", "换个会话继续这个任务".
mentor-report-html
Generate a clean, light-theme, Georgia-serif, SINGLE self-contained HTML report that shows a system or codebase as a large interactive flowchart: each node is a module / step / code unit, clicking a node opens a detail panel (file path, purpose, key functions/classes), and a knowledge-graph view highlights how those units call and depend on each other. No server, no build, no external CDN — the reviewer just opens the .html file. Use when the user says "make a visual code walkthrough", "diagram this system for my mentor / team review", "interactive architecture map", "flowchart of the pipeline", "call-graph HTML", or 中文触发词:"给导师做一份可视化代码讲解 / 架构图 / 调用关系图 / 交互式流程图 / 单文件 HTML 报告".
pr-package
Turn a working git diff into a clean, review-ready pull request: draft a strong, factual PR description and run a self-review checklist that follows common open-source conventions. Use when the user says "write a PR description", "open a pull request", "prepare/package this PR", "self-review my diff before I push", "PR checklist", or in Chinese "写 PR 描述", "提交 PR", "准备一个拉取请求", "PR 自检清单", "把 diff 整理成 PR". Community-standard scaffolding only — no house style, no proprietary review process.
verify-evidence
Evidence-first verification discipline for code changes: never claim "done", "fixed", or "working" without evidence. Identify or create a failing reproduction, fix until it passes, capture copy-pasteable verification commands with their real output, and report faithfully — if something still fails, say so and show the output. Use this whenever you are about to say a change works, close out a bug fix, hand off a result, or write a status/verification report. Trigger phrases: "verify this", "prove it works", "show me evidence", "did the fix actually work", "reproduce the bug first", "验证一下", "拿出证据", "先复现再修", "别没验证就说好了".
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.