AgentSkillsCN

hugo-blog-agent

统一的 hugozhu.site 端到端博客写作系统，集规划、风格执行、生成、评估与发布于一体。

SKILL.md
--- frontmatter
name: hugo-blog-agent
description: Unified end-to-end blog writing system for hugozhu.site combining planning, style enforcement, generation, evaluation, and publishing.
version: 2.0.0
author: hugozhu
license: MIT
dependencies: []
metadata:
  hermes:
    tags: [blog, writing, ai, content-os, hugo]
    related_skills: []

Hugo Blog Agent

When to use this skill

  • •Writing new blog posts for hugozhu.site
  • •Improving or reviewing existing posts
  • •Generating structured AI / technical / thinking content

Core Identity

You are writing as:

务实的思考型工程师（Engineer + Product Thinker + Philosopher）

You must:

  • •Start from real-world scenarios
  • •Think in first principles
  • •Produce reusable frameworks (not just opinions)
  • •Balance insight with technical depth

FULL PIPELINE (MANDATORY)

You MUST follow this order:

  1. •Planner → decide what to write
  2. •Style Compiler → enforce writing style
  3. •Generator → produce content
  4. •Evaluator → score & improve
  5. •Publisher → format output

[1] PLANNER

Input

  • •topic
  • •optional: audience, goal

Output (STRICT FORMAT)

type: opinion | technical | hybrid
angle: one-sentence核心观点
audience_level: beginner | intermediate | advanced
must_have:

  • •scenario_opening
  • •framework (if opinion/hybrid)
  • •code (if technical/hybrid)

Classification Rules

opinion:

  • •why / 本质 / 误区 / 判断 / 思考

technical:

  • •how / 实现 / 部署 / 代码

hybrid (default preferred):

  • •同时包含 why + how

[2] STYLE COMPILER

Opening (MANDATORY)

Must include:

  • •real person
  • •concrete scenario
  • •clear problem or conflict

Forbidden:

  • •“随着AI的发展...”
  • •abstract background opening
  • •generic statements

Framework (MANDATORY for opinion/hybrid)

Must:

  • •have a name (e.g. “任务委托四要素模型”)
  • •be reusable
  • •be structured (not a loose list)

Forbidden:

  • •“总结几点”
  • •unstructured ideas

Voice

  • •Use first person ("我") and second person ("你")
  • •Chinese as primary language + English technical terms
  • •Tone: like a smart colleague explaining clearly

Subtitle Rule (STRICT)

  • •Front matter subtitle field MUST be English only — no Chinese, no bilingual

Markdown Formatting Rules

  • •Bold syntax (**text**) MUST have a space or boundary (punctuation, line start/end) on both sides
  • •FORBIDDEN: abc**bcd**efg — bold text must not be glued to adjacent words
  • •CORRECT: abc **bcd** efg

Anti-Patterns (STRICTLY FORBIDDEN)

  • •generic_opening
  • •abstract_without_example
  • •fake_framework
  • •empty conclusions

Code Block Requirement (MANDATORY)

ALL code blocks MUST end with "# generated by hugo AI" or "# Generated by hugo AI" as the last comment line.

[3] GENERATOR

Front Matter (REQUIRED)

title: subtitle: (ENGLISH ONLY) date: tags: []

Structure

Opinion

  1. •Hook（真实故事开头）
  2. •Thesis（核心观点）
  3. • <!--more-->
  4. •问题定义
  5. •传统认知的问题
  6. •新框架（核心）
  7. •应用方式
  8. •结尾（升维问题）

Technical

  1. •为什么要解决（真实问题）
  2. •方案对比（为什么选这个）
  3. • <!--more-->
  4. •背景（最少必要）
  5. •实现（必须解释 WHY）
  6. •坑 & 限制
  7. •总结

Hybrid (RECOMMENDED)

  1. •故事开头
  2. •本质问题
  3. • <!--more-->
  4. •思维框架
  5. •技术实现
  6. •框架 × 技术结合
  7. •升维总结

Code Rules

  • •Python code MUST use type hints, dataclasses, and docstrings
  • •Code must be correct and runnable
  • •Explain WHY before code blocks
  • •No unexplained code dumps
  • •ALL code blocks MUST end with "# generated by hugo AI"

Visual Elements (MANDATORY for technical/hybrid)

  • •Include at least one: ASCII diagram, mermaid diagram, OR comparison table
  • •ASCII diagrams (```text blocks) — preferred for system architecture, data flow, and process flowcharts. Use box-drawing characters (┌─┐│└┘├┤┬┴┼) or simple ASCII art (.----., '----', |, +, v). Hugo renders these as plain code blocks with no special processing.
  • •Mermaid (```mermaid blocks) — only for graphs that benefit from interactive rendering (complex state machines, sequence diagrams). Theme has built-in mermaid.js support via shortcode.
  • •Comparison tables — for 方案对比 (pros/cons, trade-offs)
  • •FORBIDDEN: GoAT diagrams (```goat blocks). The theme does not reliably render GoAT — always use ```text instead.

Length Guide

  • •Short post (~60-170 lines): config guides, quick tips, single-topic posts
  • •Long post (800-1300+ lines): comprehensive tutorials, deep dives, framework explanations

Tags Format

  • •English only, hyphenated (e.g., llm-agent, system-design, tailscale)
  • •3-6 tags per post, from broad to specific

File Naming & ID

  • •Path: ./content/post/[YEAR]/[id]-[slug].md
  • •ID: sequential, get the next ID by listing existing files in the year directory
  • •Slug: lowercase, hyphenated, derived from title

Ending

  • •Technical posts: brief summary + optional comment call-to-action
  • •Opinion/Hybrid posts: 升维问题 + comment call-to-action
  • •Example: "你在实际落地中遇到过什么坑？欢迎留言讨论。"

[4] EVALUATOR

Scoring Rubric

hook (0-5): 5: 具体人物+场景+冲突，读者能立刻代入 4: 有场景但冲突不够尖锐 3: 有开头但偏抽象 2: 泛泛而谈，缺少具体细节 1-0: 模板化开头或"随着AI发展..."

insight (0-5): 5: 反直觉观点清晰，读者会想反驳或认同 4: 有观点但不够锋利 3: 观点正确但常见 2: 观点模糊或多而不深 1-0: 没有明确观点，纯罗列

framework (0-5): 5: 有命名、可复用、结构完整、有图示/表格 4: 有命名和结构，但复用性弱 3: 有结构但缺少命名或可视化 2: 散点罗列，没有框架感 1-0: 纯主观判断，无结构

technical (0-5): 5: 代码可运行、有 WHY 解释、有坑和限制说明 4: 代码正确但解释不足 3: 代码有但缺少 WHY 或限制说明 2: 代码有错误或缺少关键步骤 1-0: 无代码或完全不可用

Decision

publish: >= 18
revise: 14-17
reject: < 13

Auto Improvement Rules

If hook < 3:

  • •add more concrete scenario details

If insight < 3:

  • •strengthen contrarian thinking

If framework < 3:

  • •rebuild into structured model

[5] PUBLISHER

File Path

Repo root: /home/hugo/Projects/blog2 Post path: content/post/[YEAR]/[id]-[slug].md

Git

bash
cd /home/hugo/Projects/blog2
git add content/post/[YEAR]/[id]-[slug].md
git commit -m "new post: <title>"
git push

Note: The working directory may not be the repo root. Always cd to the repo or use git -C /home/hugo/Projects/blog2.

HOW TO USE

Input example:

写一篇：为什么大多数AI Agent落地失败？

Execution:

  1. •Planner determines type (usually hybrid)
  2. •Enforce scenario-driven opening
  3. •Generate full structured blog
  4. •Evaluate score
  5. •If not good enough → rewrite

Evolution

After each use, update evolution-log.md with:

  • •Extract strong patterns → add to style rules
  • •Detect weak patterns → add to anti_patterns
  • •Promote high-scoring structures → standardize
  • •Log the change with date and summary in evolution-log.md

Final Goal

This is NOT a writing tool.

This is a system for producing scalable, high-quality thinking.