功能描述

Obsidian second-brain and knowledge-base skill. Save any URL, article, tweet, or X post to your Obsidian vault as clean, categorized, wikilinked markdown. Tw...

使用说明 (SKILL.md)

Obsidian Librarian

Name: Obsidian Librarian
Author: shahalay007

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

Stage the raw source in _Inbox/.
Run Gemini pass 1 to clean and structure it into markdown.
Run Gemini pass 2 to choose category, tags, source attribution, and candidate wikilinks.
Scan existing vault notes for titles and aliases to resolve [[wikilinks]].
Write the final note with YAML frontmatter into the chosen category folder.
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 \x3CCategory>, --threshold \x3Cfloat> (default 0.65), --limit \x3CN> (default 5).

Reindex The Vault

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

Add --file \x3Cpath> 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.

安全使用建议

This skill appears to do what it says: it will write cleaned markdown notes into the Obsidian vault path you supply and use Gemini for processing and RAG. Before installing, consider: 1) The skill will send content to Gemini (GEMINI_API_KEY) for both text synthesis and embeddings — only provide a key you trust and understand the provider's data usage. 2) URL ingestion is routed through Apify by design (APIFY_API_KEY required for that path), so any web pages you capture will be processed by Apify; avoid providing APIFY_API_KEY if you want strictly local fetching. 3) Supabase storage is optional but will store embeddings/content on the supplied SUPABASE_URL if enabled — only supply credentials for databases you control. 4) The skill will write and (by default) delete staged inbox files inside your vault; ensure OBSIDIAN_VAULT_PATH points to a containerized or intended location. 5) If you need higher privacy, avoid providing APIFY or SUPABASE credentials and review/host the scripts yourself. If you want more assurance, run the Python files in a sandbox and inspect vault outputs and network traffic during a test ingest.

功能分析

Type: OpenClaw Skill Name: obsidian-librarian Version: 0.2.7 The obsidian-librarian skill is a legitimate and well-structured tool designed to automate the ingestion of web content (URLs, tweets, text) into an Obsidian vault and provide RAG-based querying. It utilizes Gemini for content synthesis, Apify for web scraping, and supports both local JSON and Supabase pgvector for vector storage. While the skill requires significant permissions—including network access for API calls and file system access to the Obsidian vault—its behavior is entirely consistent with its stated purpose. Security-sensitive operations, such as the use of `subprocess.run` for `curl` in `scripts/common.py`, are implemented using argument lists rather than shell execution, reducing the risk of injection. No evidence of malicious intent, data exfiltration to unauthorized endpoints, or prompt injection was found.

能力标签

requires-sensitive-credentials

能力评估

✓ Purpose & Capability

Name/description (Obsidian librarian) matches the code and runtime instructions: the skill reads/stages inputs, runs a two‑pass Gemini pipeline, resolves wikilinks against the vault index, writes markdown notes into the vault, and optionally indexes to Supabase. Required binaries (python3, curl) and required env vars (GEMINI_API_KEY, OBSIDIAN_VAULT_PATH) are appropriate for these operations.

ℹ Instruction Scope

SKILL.md and the scripts instruct the agent to read/write files inside the configured vault inbox and to call external services (Gemini for generation/embeddings, Apify for URL scraping). The instructions are focused on the stated task but do mandate 'Always use Apify' for URL fetches — that routes fetched page content through an external service. URL ingestion will not be local unless you avoid that path.

✓ Install Mechanism

No install spec or remote download is present; the skill is shipped as Python scripts and run via python3. No high‑risk install steps (no external archives or shortener URLs) were observed.

ℹ Credentials

Primary credential GEMINI_API_KEY is reasonable and required for both ingest and RAG. OBSIDIAN_VAULT_PATH is required and justified. APIFY_API_KEY and SUPABASE_* are optional for URL ingestion and Supabase-backed RAG respectively — they are not required by default but will enable external transmission/storage of data if provided. The number and type of env vars requested are proportionate to the feature set, but supplying optional keys enables external data flow.

✓ Persistence & Privilege

Skill does not request always:true or any elevated platform privilege. It writes and deletes files only within the configured vault/inbox and its own index path (or to your Supabase if configured). It does not modify other skills or global agent settings.

版本历史

v0.2.7

Sync SKILL.md frontmatter version with _meta.json

v0.2.6

Sharper description for search; second-brain framing; Supabase setup notes; dead-config cleanup

v0.2.3

Preserve X post captures with deterministic synthesis for X status URLs instead of generic article-style rewrites.

v0.2.2

Improve URL ingest: keep Apify-first reads for all URLs, follow linked URLs from X posts, and reject placeholder X article pages.

v0.2.1

Trimmed bundle: removed unused dev test script and added publish guard for test files.

v0.2.0

Initial public release: two-pass Gemini ingest + local RAG (cosine over JSON or Supabase pgvector).

元数据

Slug obsidian-librarian

版本 0.2.7

许可证 MIT-0

累计安装 0

当前安装数 0

历史版本数 6

常见问题