360 Secure SkillHub
360 安全技能中心

Obsidian Librarian

面向 Obsidian 知识库的 Skill:接收粘贴文本、本地文件或 URL,经两遍 Gemini 处理完成结构化、分类、打标签与 wikilink 解析后写入分类文件夹;并可对整个仓库做 RAG 问答返回带引用的答案,索引可用本地 JSON 或 Supabase pgvector,URL 读取依赖 Apify。

shahalay007
shahalay007
热度 54.47
最近更新 2026-05-11 17:44:07
分类:知识依赖:代码执行 / 读写文件 / 第三方 API / 内部调用大模型 / 联网浏览搜索应用场景:笔记整理 / 摘要总结 / 研究 / 工作流

安装技能

复制下面的话术发给你的 Agent,它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill: npx secure-skills-cli add obsidian-librarian
Metadata
name
obsidian-librarian
description
Obsidian second-brain and knowledge-base skill. Save any URL, article, tweet, or X post to your Obsidian vault as clean, categorized, wikilinked markdown. Two-pass Gemini pipeline handles structure, tags, and categories. Ask your whole vault anything with RAG, backed by a local JSON index or Supabase pgvector. URL fetch via Apify. Triggers on "save this", "save it", "save this url", research capture, vault search, or querying saved notes.
version
0.2.7
metadata
{"openclaw":{"primaryEnv":"GEMINI_API_KEY","requires":{"env":["GEMINI_API_KEY","OBSIDIAN_VAULT_PATH"],"bins":["python3","curl"]},"homepage":"https://github.com/openclaw/obsidian-librarian"}}

Obsidian Librarian

A second brain for Obsidian, on autopilot. Drop any URL, article, tweet, X post, or pasted text into OpenClaw and it lands in your vault as a clean, categorized, wikilinked markdown note. Then ask your whole vault anything and get grounded answers with citations.

Use this skill when the user wants OpenClaw to store text or a URL in the Obsidian vault as a cleaned, categorized markdown note, or to search and query notes they have already saved.

Trigger shortcuts:

  • Treat save this, save it, save this url, and save this link as Obsidian-librarian requests when the same message contains a URL, pasted text, or quoted content to preserve.
  • Treat short follow-ups like save it as Obsidian-librarian requests when the immediately preceding user message provided the text or URL to store.
  • Treat phrases like search my notes, search my vault, search Obsidian, what do my notes say about ..., ask my vault, and query my saved notes as Obsidian-librarian requests that should run the RAG ask path.
  • If the message only says save this or save it with no actual content or URL available in context, do not guess; ask what should be saved.
  • If the intent is ambiguous between saving to the local filesystem versus saving to the knowledge vault, prefer the Obsidian vault when the content looks like a note, article, research snippet, or social post.

The vault is mounted in the container at /data/.openclaw/obsidian-vault. Raw inputs are staged in /data/.openclaw/obsidian-vault/_Inbox, then processed into category folders.

Environment

Required:

  • GEMINI_API_KEY: Gemini API key used for both ingest and RAG answer generation.
  • OBSIDIAN_VAULT_PATH: Absolute path to the mounted Obsidian vault.

Conditional:

  • APIFY_API_KEY: Required for URL ingestion.

Optional:

  • OBSIDIAN_INBOX_FOLDER: Override the inbox folder name. Default: _Inbox.
  • OBSIDIAN_GEMINI_MODEL: Primary model override for librarian operations.
  • GEMINI_MODEL: Fallback model name when OBSIDIAN_GEMINI_MODEL is unset.
  • OBSIDIAN_RAG_INDEX_PATH: Override the local JSON RAG index path.
  • SUPABASE_URL: Enable Supabase-backed vector storage.
  • SUPABASE_KEY: Supabase API key for vector storage.
  • EMBEDDING_MODEL: Embedding model override. Default: gemini-embedding-001.
  • EMBEDDING_DIMENSIONS: Embedding size. Default: 384.

URL handling policy:

  • Always use Apify to read the URL first.
  • For x.com / twitter.com post URLs, use the dedicated Apify tweet actor.
  • If an X post contains linked URLs, follow those linked URLs through the same Apify-first path before falling back.
  • If direct URL reading fails, run a web-search fallback and stage the search-result snapshot instead.
  • If both stages fail, surface the full error back to OpenClaw instead of silently swallowing it.

Supported Inputs

  • Pasted text
  • A local text/markdown file
  • A blog/article URL
  • An existing file already sitting in _Inbox
  • A natural-language question about the saved vault

Workflow

  1. Stage the raw source in _Inbox/.
  2. Run Gemini pass 1 to clean and structure it into markdown.
  3. Run Gemini pass 2 to choose category, tags, source attribution, and candidate wikilinks.
  4. Scan existing vault notes for titles and aliases to resolve [[wikilinks]].
  5. Write the final note with YAML frontmatter into the chosen category folder.
  6. Delete the _Inbox file only after the final note is written successfully.

Ingest From Text File

python3 {baseDir}/scripts/run_pipeline.py ingest --text-file /data/.openclaw/workspace/input.txt

Ingest From URL

python3 {baseDir}/scripts/run_pipeline.py ingest --url "https://example.com/article"

Ingest An Existing Inbox File

python3 {baseDir}/scripts/run_pipeline.py ingest --inbox-file /data/.openclaw/obsidian-vault/_Inbox/some-file.md

Ask The Vault (RAG)

python3 {baseDir}/scripts/run_pipeline.py --vault-path /data/.openclaw/obsidian-vault ask "What do my notes say about AI agents?" --print-json

Optional flags: --category <Category>, --threshold <float> (default 0.65), --limit <N> (default 5).

Reindex The Vault

python3 {baseDir}/scripts/run_pipeline.py --vault-path /data/.openclaw/obsidian-vault reindex

Add --file <path> to re-embed a single note instead of the full vault.

Notes

  • For long pasted text, prefer writing it to a temp file under /data/.openclaw/workspace/ and using ingest --text-file.
  • Use --title "Custom Title" on ingest for an explicit note title override.
  • Use --keep-inbox only when debugging. Normal behavior is to clean up the staged source after success.
  • X status URLs preserve deterministic post metadata and captured post content instead of relying on a generic article-style rewrite.
  • The pipeline does forward-linking only in v1. Existing notes are not modified.
  • URL ingestion requires APIFY_API_KEY in the container environment.
  • RAG indexing runs after successful ingests. By default it uses a local JSON index; set SUPABASE_URL and SUPABASE_KEY to use Supabase pgvector instead (requires EMBEDDING_DIMENSIONS=384 to match sql/vault_chunks.sql).
  • SUPABASE_URL must point at a Supabase-compatible API surface. All requests are issued against /rest/v1/..., so self-hosted PostgREST needs a gateway or reverse proxy that serves that prefix.
  • Before enabling Supabase, apply sql/vault_chunks.sql to the target database. It provisions the vault_chunks table, the HNSW index, and the match_vault_chunks RPC that the ask command calls.