菜单
Initialize a new TwinMind vault in the current directory. Creates TwinMind.md configuration and vault directory structure from templates. Use when starting a fresh knowledge vault or setting up TwinMind in a new project.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | setup |
| description | Initialize a new TwinMind vault in the current directory. Creates TwinMind.md configuration and vault directory structure from templates. Use when starting a fresh knowledge vault or setting up TwinMind in a new project. |
| license | MIT |
Initialize a new TwinMind vault in the current working directory.
Before proceeding, check for existing installations:
.claude/skills/tm-* directories exist → Warn: "Detected old-style TwinMind installation (.claude/skills/tm-*). This plugin-based setup is separate. Consider removing old files to avoid confusion."<vault_dir>/ already exists with content. Setup will not overwrite existing vault data. Only missing structure files will be created."If all checks pass or user confirms, proceed.
Ask the user for customization (or accept defaults):
vaultzh-TWSetup needs the plugin's absolute path to copy templates from. Read <cwd>/.claude/twinmind/config.json with the Read tool and use its pluginRoot field as <plugin-root> in the steps below.
Cold-start fallback: if <cwd>/.claude/twinmind/config.json does not exist, the SessionStart hook has not run for this vault yet (TwinMind.md is required to trigger it). In that case:
<cwd>/TwinMind.md with minimal frontmatter (just vault_dir: vault\nvault_name: TwinMind\nlocale: zh-TW).exit 退出 Claude Code，然後 claude 重新啟動 session (讓 SessionStart hook 寫入 plugin 路徑)，再次執行 /setup 完成餘下步驟。"<plugin-root>/templates/TwinMind.md to <cwd>/TwinMind.md.vault_dir, locale, domains).<plugin-root>/templates/vault/ directory structure to <cwd>/<vault_dir>/:
System/, Cards/, Sources/, Atlas/, PARA/Inbox/, PARA/Actions/, PARA/Projects/, PARA/Archive/, PARA/Areas/, PARA/Tasks/.System/vault-index.json, Home.md, PARA/Dashboard.md..gitignore in <cwd> if not exists, listing both .obsidian/ and .claude/twinmind/ (the latter holds plugin-generated shims that regenerate on every session start).Report:
<vault_dir>/TwinMind action engine — manages standalone actions that aren't tied to any project. Use this skill when the user describes something they want to do that has a clear scope but doesn't belong to a project. Triggers: '建立行動', 'create action', '新增 action', '完成 action', 'action done', '做完了', '列出 actions', 'list actions', '升格為專案', 'promote to project', or when the user describes an executable goal like '研究 X', '設定 Y', '整理 Z'. Also use when the user says an action is getting too big and needs to become a project. Important: if the user mentions a specific project name, route to twinmind:project instead — that skill handles project-scoped actions.
TwinMind card engine. Use this whenever the user shares a thought, idea, knowledge, question, or source reference that should be captured as a card. Also handles card updates and deletions. Covers the full lifecycle: type classification, duplicate detection, card creation, automatic link inference to existing cards, and index updates. If the user says something that looks like knowledge worth remembering — this is the skill to use. For vague or unformed thoughts, route to twinmind:inbox instead.
TwinMind area engine — manages Areas of ongoing responsibility (long-term focus domains with no end date). Use this skill when the user talks about a persistent life/work domain they want to track, like '健康管理', '職涯發展', '財務規劃', or when they want to group projects and cards under a theme. Triggers: '建立領域', 'create area', '新增 area', '停用 area', 'deactivate', '重新啟用', 'reactivate', '關聯到 area', 'link to area', '列出 areas', 'list areas', '關注領域', or any mention of ongoing responsibilities, life domains, or thematic grouping of projects. Important: Areas are different from Projects (which have end dates) and from card domains (which are metadata tags). An Area is a conscious commitment to track a life domain.
TwinMind link engine for manual connections. Use when the user explicitly asks to link or unlink two specific cards (e.g. '把 X 和 Y 連起來', 'link X to Y', 'disconnect X from Y'). Handles relationship type classification, bidirectional link writing, and index updates. Not for automatic link inference during card creation — that's handled by twinmind:capture.
TwinMind inbox engine — the safety net for half-formed thoughts. Use this skill whenever the user shares something vague, unstructured, or not yet ready to become a Card, Action, Task, or Project. Classic triggers: '突然想到...', '隨手記一下', '有個想法但還不確定', 'inbox', '升格', 'promote', 'dismiss', '清理 inbox', 'triage', '待處理'. Also use this when the user's input is ambiguous — if you can't tell whether it's knowledge, a task, or an action, it belongs in the Inbox. The key question: does the user know what this should become? If not → Inbox. Important: do NOT use this for inputs that are already clear — a fact goes to twinmind:capture, '買牛奶' goes to twinmind:task, a scoped action goes to twinmind:action.
TwinMind shared post-operation pipeline — DEPRECATED as LLM subagent. Post-op is now handled programmatically via scripts/post-op.mjs. This skill file is retained for reference only.