Note to Blog

从 Obsidian Note 仓库中筛选适合发布的笔记，评估适配性，批量选题，双通道处理（快速转换 / 深度研究），并行 Agent 派发。

Prerequisites

Tool	Type	Required	Install
Python 3	cli	Yes	Pre-installed on macOS
PyYAML	pip	Yes	`pip install pyyaml`
writing-proofreading	skill	No	Included in `npx skills add niracler/skill`

Do NOT proactively verify these tools on skill load. If a command fails due to a missing tool, directly guide the user through installation and configuration step by step.

Script Location

All deterministic operations are handled by the Python script:

text

scripts/note-to-blog.py  (collect / convert / state subcommands)

Path configuration is in user-config.md.

Workflow Overview

text

Phase 1          Phase 2         Phase 3            Phase 4              Phase 5
 Collect   ──▶   Evaluate  ──▶   Interact    ──▶   Execute        ──▶   Summary
 (script)        (LLM)          (user)             (Agent Teams)         (report)
                                 ├─ select          ├─ Fast track
                                 ├─ skip            └─ Deep track
                                 └─ assign track

Phase 1: Collect

Run the

collect

script to gather all data in one call:

bash

python3 scripts/note-to-blog.py collect \
  --note-repo "~/Library/Mobile Documents/iCloud~md~obsidian/Documents/Note/" \
  --blog-content "repos/bokushi/src/content/" \
  --project-paths \
    "~/.claude/projects/-Users-sueharakyoko-code-nini-dev" \
    "~/.claude/projects/-Users-sueharakyoko-code-nini-dev-repos-bokushi" \
  --history-file "~/.claude/history.jsonl"

The script outputs a single JSON object to stdout containing:

```
candidates
```
: all eligible notes with title, summary, char_count, outgoing_links
```
clusters
```
: wikilink hub nodes (3+ inbound links) with related notes
```
published_posts
```
: existing blog posts with title, tags, collection
```
session_keywords
```
: recent Claude Code session activity signals
```
stats
```
: total_scanned, filtered_out, candidates_count

Read the JSON output and proceed to Phase 2.

Phase 2: Evaluate

Make a single LLM evaluation using the prompt template from scoring-criteria.md.

Input

Construct the evaluation prompt with the

collect

JSON data:

```
candidates
```
: title + summary + char_count for each
```
clusters
```
: hub_title + related count + link_count for each
```
published_posts
```
: title + tags for deduplication
```
session_keywords
```
: for timeliness scoring

Output

The LLM SHALL return a JSON array of 5~8 recommendations, mixing individual notes and topic clusters:

json

[
  {
    "type": "single",
    "path": "Areas/大模型(LLM)/关于后LLM时代的代码Review.md",
    "title": "关于后 LLM 时代的代码 Review 的看法",
    "score": 92,
    "collection": "blog",
    "effort": "小",
    "session_activity": "★★★",
    "duplicate_risk": "none",
    "reason": "结构完整、有真实案例、观点独特"
  },
  {
    "type": "cluster",
    "hub_title": "优雅的哲学",
    "hub_path": "Areas/生活(Life)/优雅的哲学-v2.0.md",
    "related_count": 9,
    "score": 88,
    "collection": "blog",
    "effort": "大",
    "theme_summary": "关于如何优雅地生活的哲学思考，散落在多篇笔记中",
    "reason": "主题深度足够，需要整合多篇笔记"
  }
]

If the LLM response is not valid JSON, retry once with explicit format instructions.

Phase 3: Interact

Step 3.1: Present recommendations

Display the recommendation list as a mixed table:

text

#  类型    标题                    适配分  目标   工作量  活跃   重复风险
1  单篇    后LLM时代代码Review      92    blog    小    ★★★    无
2  主题簇  优雅的哲学 (9篇关联)      88    blog    大    ★      无
3  单篇    SSH私钥加密              85    til     小    ─      无
4  单篇    Feed内容阅读姿势         82    blog    小    ★      无
...

For cluster entries, show the hub title and related note count.

Step 3.2: User actions

The user can perform batch operations:

Action	Example	Effect
Select + assign track	"1 和 3 快速转换，2 走深度"	Queue items with track assignment
Override collection	"1 放 til"	Change target collection
Batch skip	"4~6 跳过，reason: private"	Mark as skipped via `state skip`
See more	"还有别的吗"	Request additional recommendations (exclude previously shown items)
Check status	"状态"	Run `state status` to show drafted/published/skipped counts

On skip: run the state script immediately for each skipped item:

bash

python3 scripts/note-to-blog.py state skip "<path>" --reason "<reason>" \
  --note-repo "~/Library/Mobile Documents/iCloud~md~obsidian/Documents/Note/"

Step 3.3: Track assignment

Each selected item must be assigned a track:

Track	When to use	What happens
Fast (快速)	Independent, mostly complete notes	Script converts → Agent reviews → draft
Deep (深度)	Topic clusters or rough notes needing research	Agent reads all related notes → research report

Default suggestions:

```
type: "single"
```
with
```
effort: "小"
```
→ suggest fast track
```
type: "cluster"
```
or
```
effort: "大"
```
→ suggest deep track
User always makes the final decision

Step 3.4: Confirm selections

Before proceeding, display a summary:

text

确认选择：
  Fast track:
    1. 后LLM时代代码Review → blog/
    3. SSH私钥加密 → til/
  Deep track:
    2. 优雅的哲学 (9篇关联) → blog/

开始处理？

Wait for user confirmation.

Phase 4: Execute (Agent Teams)

Dispatch N parallel Agents using the Task tool, one per selected item. Each Agent operates independently.

Parallel dispatch

text

总编 (Main Agent)
├── Task Agent 1: 文章 A (fast track)
├── Task Agent 2: 文章 B (fast track)
└── Task Agent 3: 主题簇 C (deep track)

Use the Task tool to launch all Agents in a single message (parallel execution). Each Agent should be a

general-purpose

subagent with a detailed prompt containing all the information it needs (note path, script path, target collection, etc.).

Fast Track Agent Instructions

Each fast track Agent receives these instructions:

Step 1 — Run the convert script (use the absolute path to the script in the Agent prompt):

bash

python3 "<absolute-path-to-skill>/scripts/note-to-blog.py" convert "<full-path-to-note>"

Step 2 — Review the converted output (the script automatically handles):

Obsidian fields stripped (aliases, date, modified, etc.)
Frontmatter set to: title, pubDate (today), tags (merged from frontmatter + inline #tags), hidden: true
Wikilinks → plain text, image embeds → standard markdown, callouts → bold, highlights → bold, comments → removed
Verify no conversion artifacts or TODO comments from unrecognized syntax

Step 3 — Generate a description: Write a one-sentence description for the blog post frontmatter (in the same language as the article content).

Step 4 — Write the draft: Add the

description

field to the converted frontmatter, then write to

repos/bokushi/src/content/<collection>/<slug>.md

where

<slug>

is a kebab-case version of the title.

Step 5 — Return a result object:

json

{
  "status": "done",
  "note_path": "<relative note path>",
  "draft_path": "<collection>/<slug>.md",
  "description": "<generated description>",
  "issues": ["<any issues found>"],
  "suggestions": ["<improvement suggestions>"]
}

Deep Track Agent Instructions

Each deep track Agent receives these instructions:

Step 1 — Read all related notes: Read the hub note and all related notes listed in the cluster.

Step 2 — Produce a structured research report:

markdown

## 主题报告：<hub_title>

### 涉及笔记 (N 篇)
- <title> (hub, <char_count>字)
- <title> (<char_count>字)
- ...

### 主题地图
- 核心论点：...
- 子话题 A：...（涉及 N 篇）
- 子话题 B：...（涉及 N 篇）

### 重叠与矛盾
- <specific overlaps or contradictions found>

### 缺口
- <missing arguments, incomplete sections>

### 建议大纲
1. 引言：...
2. 第一部分：...（来源：笔记 A、B）
3. 第二部分：...（来源：笔记 C、需要补充）
4. 结论：...

Step 3 — Return a result object:

json

{
  "status": "done",
  "note_path": "<hub note path>",
  "related_paths": ["<related note paths>"],
  "report": "<full research report markdown>",
  "suggested_collection": "<blog/til/monthly>"
}

Important: State updates

Individual Agents do NOT update

.note-to-blog.json

directly. After all Agents complete, the main agent collects results and runs state updates sequentially:

For each fast track result:

bash

python3 scripts/note-to-blog.py state draft "<note_path>" \
  --target "<collection>/<slug>.md" \
  --note-repo "~/Library/Mobile Documents/iCloud~md~obsidian/Documents/Note/"

Deep track items are NOT marked as drafted (they need further user decision).

Phase 5: Summary

After all Agents complete, present a unified summary:

Fast track results

text

Fast Track 完成：
  ✓ 后LLM时代代码Review → repos/bokushi/src/content/blog/llm-code-review.md
    - 转换正常，无问题
    - 建议：可补充最新的 AI code review 工具对比
  ✓ SSH私钥加密 → repos/bokushi/src/content/til/ssh-key-encryption.md
    - 发现 1 个 TODO 标记需要手动处理

Deep track results

text

Deep Track 完成：
  📋 优雅的哲学 (9篇关联)
    - 研究报告已生成
    - 建议大纲：4 部分，来源涵盖 7 篇笔记
    - 下一步？
      a) 按大纲写作（调用 Agent 生成初稿）
      b) 修改大纲
      c) 暂不处理

State update confirmation

text

状态更新：
  drafted: 2 篇
  下次 collect 时这些笔记将不再出现在候选列表中。

草稿均为 hidden: true，需要手动 review 后改为 false 发布。
建议使用 /writing-proofreading 进行审校。

发布后运行 state publish 更新状态：
  python3 scripts/note-to-blog.py state publish "<note_path>" --note-repo "..."

Detailed References

Path configuration: user-config.md
LLM evaluation prompt and scoring: scoring-criteria.md

note-to-blog

NPX Install

Tags

SKILL.md Content

Note to Blog

Prerequisites

Script Location

Workflow Overview

Phase 1: Collect

Phase 2: Evaluate

Input

Output

Phase 3: Interact

Step 3.1: Present recommendations

Step 3.2: User actions

Step 3.3: Track assignment

Step 3.4: Confirm selections

Phase 4: Execute (Agent Teams)

Parallel dispatch

Fast Track Agent Instructions

Deep Track Agent Instructions

Important: State updates

Phase 5: Summary

Fast track results

Deep track results

State update confirmation

Detailed References