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

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.

When NOT to Use

• 笔记库少于 5 篇时，手动选题更快
• 只想转换单篇已确定的笔记 — 直接运行 <skill-dir>/scripts/note-to-blog.py convert "<path>"
• 博客草稿已存在，只需校对 — 使用 writing-proofreading

Script Location

All deterministic operations are handled by the Python script:

<skill-dir>/scripts/note-to-blog.py  (collect / convert / state subcommands)

Path configuration is in user-config.md. All bash examples below use <PLACEHOLDER> — replace with values from user-config.md.

Workflow Overview

Step 1           Step 2            Step 3              Step 4             Step 5
 Collect   ──▶   Level Select ──▶  By Level     ──▶   Execute      ──▶   Summary
 (script)        (user)            ├─ L1 浏览          (Agent Teams)       (report)
                                   ├─ L2 推荐
                                   └─ L3 深探
                                        │
                                   Interact ──▶ track assign ──▶ confirm

Step 1: Collect

Run the collect script with paths from user-config.md:

python3 <skill-dir>/scripts/note-to-blog.py collect \
  --note-repo "<NOTE_REPO>" \
  --blog-content "<BLOG_CONTENT>" \
  --project-paths <PROJECT_PATHS> \
  --history-file "<HISTORY_FILE>"

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 Step 2.

Step 2: Level Selection

Display data volume and offer Level choice:

collect 完成：
  候选笔记: {candidates_count} 篇 (总扫描 {total_scanned}, 过滤 {filtered_out})
  主题簇: {clusters_count} 个 (3+ 引用 hub)

可选深度:
  Level 1 浏览    直接展示候选列表，手动选择          0 额外 token
  Level 2 推荐    LLM 评估 + 主题簇分析              ~2k token     ← 推荐
  Level 3 深探    Level 2 + 读取 hub 笔记全文         ~5k+ token

选择 Level (1-3)?

Quick Reference

Recommendation logic

用户明确说「想发现主题」「有什么可以整合的」时 → 推荐 Level 3。

Step 3: By Level

Level 1: Browse

Skip LLM evaluation. Display candidates sorted by char_count descending:

#  标题                         字数    链接数
1  关于后LLM时代的代码Review     3200    5
2  SSH私钥加密                   1200    2
3  Feed内容阅读姿势              1800    3
...

User selects items by number. All selections go to fast track only (Level 1 does not offer deep track).

After selection, skip to Confirm & Execute below.

Level 2: Recommend

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

Input: Construct the evaluation prompt with collect JSON data (candidates, clusters, published_posts, session_keywords).

Output: The LLM SHALL return a JSON array of 5-8 recommendations. See scoring-criteria.md for the full specification.

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

After evaluation, proceed to Interact below.

Level 3: Deep Explore

Same as Level 2, but before calling the LLM, Read the full text of each cluster hub note and append it to the evaluation prompt.

For each cluster in the collect JSON where hub_path is not null:

1. Read the hub note full text from the Note repository
2. Append it to the LLM prompt under a ## Hub 笔记全文 section (see scoring-criteria.md for the Level 3 input format)

This gives the LLM actual content context for cluster recommendations instead of just metadata.

After evaluation, proceed to Interact below.

Interact (Level 2/3)

Present recommendations

Display the recommendation list as a mixed table:

#  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
...

User actions

On skip: run immediately:

python3 <skill-dir>/scripts/note-to-blog.py state skip "<path>" --reason "<reason>" \
  --note-repo "<NOTE_REPO>"

Track assignment

Default: effort: "小" → fast; type: "cluster" or effort: "大" → deep. User decides.

Confirm & Execute

Display a confirmation summary (all Levels):

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, then dispatch.

Parallel dispatch

Dispatch N parallel Agents using the Task tool, one per selected item.

其他 Agent 环境：以下 Fast/Deep track 任务相互独立，可按顺序依次执行。

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. Each Agent should be a general-purpose subagent with a detailed prompt containing all the information it needs.

See agent-instructions.md for the complete Fast Track and Deep Track agent prompt templates.

State updates

Individual Agents do NOT update .note-to-blog.json directly. After all Agents complete, the main agent runs state updates sequentially:

python3 <skill-dir>/scripts/note-to-blog.py state draft "<note_path>" \
  --target "<collection>/<slug>.md" \
  --note-repo "<NOTE_REPO>"

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

Summary

After all Agents complete, present a unified summary:

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
  ✓ 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 完成：
  📋 优雅的哲学 (9篇关联)
    - 研究报告已生成
    - 下一步？ a) 按大纲写作  b) 修改大纲  c) 暂不处理

状态更新：
  drafted: N 篇

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:
  python3 <skill-dir>/scripts/note-to-blog.py state publish "<note_path>" --note-repo "<NOTE_REPO>"

Detailed References

• Path configuration: user-config.md
• LLM evaluation prompt and scoring: scoring-criteria.md
• Agent prompt templates: agent-instructions.md