Source Library

通过本地 Node 脚本把用户分享的链接保存为结构化 markdown 摘要，包含关键论点、引文、分析、标签与衰减标记，并提供搜索、列表、统计、连接映射、冲突检测、阅读队列与批量导入命令，强调交叉引用与长期可检索。

don-gbot

热度 60.62

最近更新 2026-05-17 12:29:46

分类：知识依赖：代码执行 / 读写文件应用场景：研究 / 笔记整理 / 摘要总结 / 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 source-library

Metadata

name: source-library
description: Searchable knowledge base that captures and cross-references everything users share. Auto-triggers when user shares ANY URL (article, tweet, thread, repo, video, paper). Saves structured summaries with key claims, quotes, analysis, tags, and decay tracking. Cross-references sources, maps connections, detects conflicts, and manages reading queue. Triggers on: shared URLs, "source library", "what have I read", "search sources", "find that article about", "remember when I shared", "conflicts", "connections". Do NOT use for general web browsing, bookmark management, or fetching pages without saving.
allowed-tools: Bash(node:*)
compatibility: Requires Node.js 18+. Uses local markdown search for retrieval. No API keys needed. No external dependencies. Works on Linux/macOS.
metadata: {"author":"DaDefiDon","version":"2.0.0","category":"knowledge-management","tags":["sources","research","knowledge-base","cross-reference"]}

Source Library

A persistent, searchable knowledge base built from everything the user shares. Not a bookmark manager — a cross-referenced memory system with connection mapping, conflict detection, and confidence decay.

Quick Start

node scripts/source-library.js setup — creates directories
Share any URL in chat — the agent auto-processes and saves it
Use node scripts/source-library.js search "query" to find past sources

Auto-Trigger Behavior

When the user shares any URL, without being asked:

Search first — use local source search to find related existing sources. Surface specific connections.
Analyze with context — discuss the new source with existing knowledge layered in.

Save with substance:

node scripts/source-library.js save --name "Title" --url "https://..." --author "Name" --type "article" --tags "topic1, topic2" --claims "Claim 1. Claim 2." --analysis "Why this matters" --context "How it came up"

Every entry must be useful months later without re-reading the original.

Command Reference

All commands via node scripts/source-library.js <command>:

Command	Description
`setup`	Create directories, first-run welcome
`save --name "..." --url "..." [--author --type --tags --summary --claims --analysis --context --slug --related --decay --date --force]`	Save a source
`list [--type tweet] [--tag crypto] [--decay]`	List sources with optional filters
`search "query" [--limit 10]`	Search local source summaries with relevance scoring
`stats`	Library statistics (total, by type, by tag, disk usage)
`connections [--clusters\|--orphans]`	Map relationships between sources
`conflicts`	Detect contradictions via sentiment heuristics
`queue add "url" [--note "..."]`	Add URL to reading queue
`queue list`	Show queued items
`queue next`	Show oldest unprocessed item
`queue done "url-or-index"`	Remove from queue
`teach "topic" [--limit 20]`	Synthesize knowledge from related sources
`import file.json`	Bulk import from JSON (full objects or URL array)

Entry Format

Each source lives at life/source/{slug}/summary.md:

# Title

**Source:** URL
**Author:** Name (@handle)
**Date:** YYYY-MM-DD
**Type:** tweet|thread|article|repo|video|paper
**Tags:** comma-separated
**Decay:** fast|normal|slow

## Key Claims
- Actual arguments, mechanics, data points
- Specific enough to be useful without re-reading original

## Notable Quotes
- Direct quotes worth remembering verbatim

## Analysis
What matters. Connections to other knowledge. Why it's significant.

## Context
Why it was shared. Decisions made based on this.

## Related Sources
- [[other-source-slug]]

Quality Rules

No vibes. "Interesting macro take" is worthless. Capture specific claims, mechanics, data.
Capture mechanics, not conclusions. Store the reasoning, not just the takeaway.
Include quotes. Direct quotes are highest-fidelity knowledge.
Tag generously. Tags enable future discovery. Include topic, author, domain, entities.
Record decisions. If a source led to action, capture that in Context.
Cross-reference. A connected source is knowledge; an isolated source is a bookmark.

Do NOT Use For

General web browsing or page fetching without saving
Bookmark management (no "save for later" without substance)
Summarizing pages the user didn't ask to save
Anything outside the user's knowledge base