connect-the-dots

name	connect-the-dots
description	Knowledge orchestration for Obsidian vault. Use when: daily scan of new KB articles, updating MOC files, maintaining Vault Index, connecting new knowledge to existing, or evaluating article relevance. Runs as daily cron and on-demand when Park sends research requests.
platform	both
scope	agent:wendy

Connect the Dots

Knowledge orchestration skill — scan new content, judge relevance, connect to existing knowledge, maintain MOC structure, update Vault Index.

Core Principle

You are the brain, not the librarian. Monica collects and files. You connect, judge, and produce insight. Every new piece of knowledge must be woven into the existing web, or explicitly discarded.

Vault Structure

~/work/agents-co/knowledge-graph/                    ← Obsidian vault
├── _system/                         ← L3: Principles (always loaded)
├── MOC/                             ← Map of Content (hub nodes)
│   ├── MOC-AI-Agent.md              ← Main hub → 5 sub-MOCs
│   ├── MOC-AI-Memory.md
│   ├── MOC-AI-Workflow.md
│   ├── MOC-AI-MultiAgent.md
│   ├── MOC-AI-Skills.md
│   ├── MOC-AI-Security.md
│   ├── MOC-Trading.md               ← Polymarket / CN / US / Crypto / Commodities
│   └── MOC-Media.md                 ← Content creation / Business
├── research/                        ← Monica's output (individual articles)
├── people/                          ← People profiles
├── tools/                           ← Tool evaluations
├── vault-index.md                   ← One-line summary per note (for fast scan)
└── briefings/                       ← Daily raw output (L1, not in MOC)

Workflows

1. Daily Scan (cron trigger)

Run daily. Scan new articles added in last 24h.

Step 1: Find new files
  find ~/work/agents-co/knowledge-graph/research -name "*.md" -mtime -1

Step 2: For each new file:
  a. Read TL;DR + Key Takeaways + tags
  b. Determine which MOC(s) it belongs to
  c. Add article row to MOC table (title | one-line summary | relevance)
  d. Add MOC backlink to article's 连接 section
  e. Find 2-3 existing articles it connects to → add [[双向链接]] if missing
  f. Add one-line entry to vault-index.md

Step 3: Cross-pollinate
  - Any pattern emerging across today's articles? (3+ articles touching same theme)
  - If yes → note it in memory/patterns.md
  - If a new sub-theme is forming (5+ articles) → consider proposing new MOC

Step 4: Output summary (for daily distillation)
  - X new articles scanned
  - Key connections found
  - Any emerging patterns

2. On-Demand: Link Processing (Park sends URL to Wendy)

Step 1: Read content (web_fetch / summarize / whisper)

Step 2: Judge relevance (1-10)
  ≤ 4 → Tell Park "不太相关：[reason]". Stop.
  5-7 → Light note (TL;DR + 要点 + tags). Save to research/.
  8-10 → Full research note (Monica template). Save to research/.

Step 3: Connect
  a. Find 2-3 existing notes with strongest connection
  b. Add [[双向链接]] bidirectionally
  c. Add to appropriate MOC(s)
  d. Update vault-index.md

Step 4: Reply to Park
  📎 **{title}**
  要点：{one line}
  跟我们的关系：{one line}
  行动建议：{action or "存档，暂不行动"}
  已存：[[{note}]] → [[{MOC}]]

3. On-Demand: Topic Research (Park says "研究一下 X")

Step 1: Search existing KB
  grep + memory_search for related content

Step 2: Web research
  web_search → read top 5-10 sources

Step 3: Synthesize report
  - 现状概览
  - 关键玩家/观点
  - 跟 Park 的 thesis 的关系
  - Mermaid diagram (if structural relationships exist)
  - 可行动建议

Step 4: Save to research/ + connect + update MOC

Step 5: Reply with concise summary

4. On-Demand: Tool Evaluation (Park says "看看这个工具")

Step 1: Read docs / GitHub / demo

Step 2: Evaluate
  - 是什么 / 解决什么问题
  - 跟现有工具关系：替代？互补？无关？
  - 安装成本 / 学习成本
  - 建议：装 / 不装 / 观望

Step 3: Save to tools/ + update vault-index.md

Step 4: Reply with verdict + install plan if recommended

MOC Update Rules

When adding an article to a MOC:

1. Add row to the ## 文章 table: | [[article-name]] | one-line summary | X/10 |
2. If article connects to articles in OTHER MOCs → add cross-reference in ## 连接 section
3. If ## 关键洞察 needs updating based on new evidence → update it
4. Never remove existing entries, only add

Vault Index Format

~/work/agents-co/knowledge-graph/vault-index.md:

# Vault Index
<!-- One-line per note. Agent scans this first, then reads full note if needed. -->

## Research
- [[260214-ClawVault-Agent记忆架构-Obsidian式]] — Markdown > vector DB for agent memory. Vault Index pattern.
- [[260213-OpenClaw记忆管理三层架构实战]] — 热/冷/原始三层, P0/P1/P2优先级, 自动归档.
...

## Tools
- [[claudian]] — Obsidian plugin embedding Claude Code CLI in vault.
...

## People
- [[steipete]] — OpenClaw power user, SOUL.md 最佳实践.
...

Quality Gates

• No orphan notes: Every new note must link to ≥1 existing note + ≥1 MOC
• No stale MOCs: If a MOC hasn't been updated in 7 days but new articles exist → flag it
• No bloated vault-index: Each entry ≤ 15 words. It's a scan tool, not a summary.
• Relevance filter: Don't store ≤ 4/10 content. KB is curated, not comprehensive.