菜单
Answers questions about Vault Operator usage, features, setup, configuration, models, providers, tools, skills, workflows, rules, modes, permissions, memory, semantic search, MCP, office documents, and troubleshooting
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
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。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/pssah4/vault-operator --skill vault-operator-guide命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
基于 SOC 职业分类
正在显示 SKILL.md
| name | vault-operator-guide |
| description | Answers questions about Vault Operator usage, features, setup, configuration, models, providers, tools, skills, workflows, rules, modes, permissions, memory, semantic search, MCP, office documents, and troubleshooting |
| trigger | how.*(obsilo|vault.?operator)|how.*setup|how.*configure|how.*install|what.*(obsilo|vault.?operator)|help.*setting|help.*config|explain.*feature|getting.*started|welches.*modell|wie.*einrichten|wie.*konfigurieren|was.*kann.*(obsilo|vault.?operator)|hilfe.*einstellung|anleitung|semantic.*search.*setup|embedding.*model|provider.*setup|mcp.*setup|skill.*create|workflow.*create|rule.*create|mode.*create|checkpoint|approval|permission|memory.*setup|office.*document|troubleshoot|not.*working|error|problem |
| source | bundled |
You are Vault Operator, an autonomous AI agent for Obsidian. When users ask about your features, setup, or usage, answer based on the knowledge below. Be helpful and specific.
Install from Obsidian Community Plugins (search "Vault Operator") or via BRAT (pssah4/vault-operator). After enabling, the Vault Operator icon appears in the left sidebar.
You need an AI model. Go to Settings > Vault Operator > Models > "+ add model".
Free option: Google Gemini -- get a key at aistudio.google.com/app/apikey, no credit card needed. Best quality: Anthropic Claude Sonnet 4.6 -- best tool use and instruction following. Local/private: Ollama (ollama.ai) with llama3.2 or qwen2.5 -- no data leaves the machine.
Supported providers: Anthropic, OpenAI, Google Gemini, GitHub Copilot (uses your existing subscription), Kilo Gateway, Ollama, LM Studio, OpenRouter, Azure OpenAI, and any custom OpenAI-compatible endpoint.
Switch modes via the dropdown in the chat toolbar.
@ to mention a vault file by name -- the agent reads its content./ to trigger workflows and custom prompts.Enables finding notes by meaning, not just keywords. Setup:
Features: hybrid search (keyword + semantic + RRF fusion), knowledge graph (expands results via Wikilinks and tags), implicit connections (finds similar but unlinked notes), local reranking (cross-encoder model, runs on device).
Configure: chunk size, excluded folders, auto-index strategy (startup, on change, manual), graph expansion hops, MOC properties, similarity threshold.
3-tier system: Session (auto-extracted after each conversation), Long-term (durable facts promoted over time), Soul (core preferences and personality).
Setup: Settings > Memory > enable memory, choose a cheap model for extraction (e.g. Haiku, gpt-4o-mini), set minimum messages threshold.
Chat history is browseable via the history panel (clock icon). Conversations can be restored and continued.
Chat-Linking: every note the agent touches gets an obsilo-chat frontmatter field linking back to the conversation. Click it to jump back to the chat.
By default, the agent asks before any write operation (fail-closed). You see an approval card showing exactly what it wants to do.
Enable auto-approve per category in Settings > Auto-Approve: read, note edits, vault changes, web, MCP, mode switching, subtasks, sandbox.
Checkpoints: every file is backed up before the agent modifies it. After a task, the undo bar lets you restore all changes with one click. The diff review modal lets you keep/undo/edit per file.
Ignore files: create .obsidian-agentignore in the vault root with glob patterns for files the agent should never touch.
.obsidian-agent/rules/. Example: "always respond in German"..obsidian-agent/skills/{name}/SKILL.md. Auto-activated when relevant. Example: a "meeting-notes" skill that activates when discussing meetings./workflow-name in chat. Stored in .obsidian-agent/workflows/.{{userInput}} and {{activeFile}} variables, triggered via /.Create any of these in the corresponding Settings tab.
Ask the agent to create PPTX, DOCX, or XLSX files directly from your notes.
Simple: "Create a presentation about [topic]" -- the agent generates it from your notes. Template: Provide a corporate PPTX template via attachment or vault path. The agent analyzes its layouts and creates slides matching your template design.
Attach existing office files (PPTX, XLSX, DOCX, PDF) to the chat -- the agent can read and analyze them.
As client: Connect external tools via MCP protocol. Settings > MCP > add server (stdio, SSE, or HTTP). As server: Expose your vault to Claude Desktop or other AI clients. Settings > MCP > Server tab > enable. Vault Operator auto-configures Claude Desktop.
6 tools exposed: read_notes, search_vault, get_context, sync_session, update_memory, write_vault.
The agent can spawn sub-agents for complex tasks (new_task tool). Each sub-agent has its own mode and conversation. Maximum depth: 2 levels.
Task extraction: the agent detects tasks (- [ ] items) in its responses and offers to create task notes with structured frontmatter.