Claude Code 博客技能套件：30 个子技能、5 个智能体、5 道门禁的 v1.9.0 Blog Delivery Contract，针对 Google 排名和 AI 引用进行双重优化。正在 AI-Marketing-Hub/claude-blog（AI Marketing Hub Pro 社区）积极开发；公开发行版在此发布。
Claude Blog Free, open-source AI blog writer for Claude Code . Posts that are researched, sourced and reviewed before you read a word. Install · Demo · Commands · Team · Pricing · How it works · FAQ Claude Blog is a free blog writing skill for Claude Code. Give it a topic and it researches current sources, writes the post, checks the SEO, has an independent reviewer score it and runs five delivery gates. You only…
Claude Code blog skill suite: 30 sub-skills, 5 agents, 5-gate v1.9.0 Blog Delivery Contract, dual-optimized for Google rankings and AI citations. Active development at AI-Marketing-Hub/claude-blog (AI Marketing Hub Pro community); public releases ship here.
Claude Blog is a free blog writing skill for Claude Code. Give it a topic and it researches current sources, writes the post, checks the SEO, has an independent reviewer score it and runs five delivery gates. You only see the post once it passes.
What you get:
A finished post. Markdown, rendered HTML and PDF, a hero image with its license, and screenshots at phone, tablet and desktop widths plus dark mode.
A reviewer that can say no. The blog-reviewer agent scores every post out of 100 and blocks anything under 90 or with a P0 issue.
Sources you can check. Statistics come from tier 1 to 3 sources with dates, and /blog factcheck fetches every cited page to confirm the claim is there.
SEO for Google and AI answers. Answer-first structure, Schema.org markup and Google’s latest guidance, so one post works for rankings and for citations in Google AI Overviews, ChatGPT, Perplexity and Gemini.
Any language./blog multilingual writes the post, translates and localizes it, and adds the hreflang tags.
Claude Blog is built from 32 skills and 5 specialist agents, with 30 /blog commands for everything around the post: strategy, topic clusters, audits, refreshes, images and audio.
See it in action
Type /blog in Claude Code to see every command:
Write a post from one line, through research, review and all five gates:
The analyzer’s two packages install with pip install -r requirements.txt. Rendering, screenshots and generated heroes use the optional presentation extra in pyproject.toml (Patchright, WeasyPrint, Markdown, Pillow, requests and google-genai); without a renderer, gate 3 blocks delivery instead of skipping it.
Standalone installer for Unix, macOS and Windows, with digests and uninstall
Recommended: clone, review, then install. <commit-sha-or-release-tag> is a release from v2.3.0 on or a commit you reviewed, ideally a full commit SHA.
git clone https://github.com/AgriciDaniel/claude-blog.git
cd claude-blog
git checkout <commit-sha-or-release-tag>
# Review install.sh, uninstall.sh and scripts/installer_ownership.py.
chmod +x install.sh
./install.sh
Download and run on Unix and macOS, with the digest check:
Integrity is only as pinned as CLAUDE_BLOG_REF. Each digest is the SHA-256 of that one file with CRLF line endings normalized to LF, the form raw.githubusercontent.com serves (install.sh: the digest above; install.ps1: 55c40999f3d9c2261917f5160c8e880c23debc2941b1c13df4a6fc2379de180e). The digests do not cover the code that does the work: the wrapper clones the repository at CLAUDE_BLOG_REF and hands over to that revision’s own installer. A branch such as main moves, a tag can be re-pointed by whoever controls the repository, and a full commit SHA cannot. Pin a commit SHA you reviewed.
Do not install or uninstall from the v2.2.0 tag: its uninstall.sh deletes whole blog and blog-* skill folders, including files you added. The safe installer starts at v2.3.0. It never uses sudo, backs up any file it replaces to ~/.claude/claude-blog-backups/<timestamp>/, and stops with recovery steps if it finds a managed file you edited.
Uninstall with ./uninstall.sh (Unix, macOS) or .\uninstall.ps1 (Windows). More in docs/INSTALLATION.md.
Your first post:
claude # start Claude Code
/blog strategy "freelance CRM software" # find the topics worth writing
/blog write "best CRM for freelancers" # research, write, review, gates
/blog analyze posts/best-crm.md # score any post you already have
/blog multilingual "best CRM for freelancers" --languages de,fr,es
The platform is detected for you (Next.js MDX, Astro, Hugo, Jekyll, WordPress, Ghost, 11ty, Gatsby or static HTML), and frontmatter, images and schema follow it.
Commands
30 commands in six groups. Open one to see what is inside; the full reference is in docs/COMMANDS.md.
Write and shape (7 commands)
/blog write <topic>: A new post: researched, written, reviewed and gated
/blog rewrite <file>: Improve an existing post with sourced statistics (alias /blog update)
/blog outline <topic>: An outline shaped by what already ranks
/blog brief <topic>: A content brief with competitive analysis
/blog style learn <paths>: Learn your voice from 5 to 10 of your posts
/blog persona [create | list | use | show]: Writing personas with tone settings
/blog brand [init | show | update]: BRAND.md and VOICE.md that every skill reads
Plan and research (5 commands)
/blog strategy <niche>: Positioning and the topics worth writing
/blog calendar [monthly | quarterly] [niche]: An editorial calendar built on topic clusters
/blog cluster [plan | execute] <seed or cluster-plan.json>: Plan and write a pillar with its supporting posts
/blog discourse <topic>: What people ask on Reddit, YouTube and Hacker News, no API keys
/blog notebooklm ask <question>: Source-grounded answers from your NotebookLM notebooks
Check quality (6 commands)
/blog analyze <file or url>: The 100-point score in five categories
/blog seo-check <file>: On-page SEO checklist after writing
/blog factcheck <file>: Fetch every cited source and check the claim
/blog audit [directory]: Health check for a whole blog
/blog cannibalization [directory]: Posts competing for the same keyword
Native translations that keep markdown, schema and charts intact
Read, Write, Edit, Glob, Grep
Of the five agents, only the writer and the translator can edit files. The researcher is the only agent with web access, and agents cannot start other agents: /blog sends each one its job.
Who it is for:
Bloggers and solo founders who want posts they can publish without rewriting them first.
Content and SEO teams that want every post scored, sourced and checked the same way.
Agencies that write for several clients, in several languages, from one setup.
Pricing
The full comparison
Writing it yourself
Hiring a writer
AI writing apps
Claude Blog
Cost per post
Your hours
$300 to $1,000 or more
A monthly plan
Free, with your Claude Code plan
Every statistic sourced and dated
Depends
Depends
Rarely
Yes, tier 1 to 3 sources
Independent review before delivery
You
An editor, if you pay for one
No
Yes, 90 out of 100 to ship
Checked for AI answers
Rarely
Rarely
Sometimes
Yes, scored out of 15
Translations with hreflang
No
Extra cost
Sometimes
Yes, /blog multilingual
Your files stay with you
Yes
Yes
Often in their cloud
Yes, on your machine
Typical market prices, for illustration. Your numbers may differ. Generated images and narration use the paid Gemini API, and Claude Blog asks before it spends anything.
How a post ships
It researches first.blog-researcher finds current statistics from tier 1 to 3 sources, with dates, plus image leads and the gaps other posts leave open.
It writes answer first.blog-writer drafts the post with every number sourced, and /blog renders it as Markdown, HTML and PDF with a hero image.
It has to pass.blog-reviewer scores the exact draft that ships, and five gates check it. When a gate blocks, the draft is fixed, rendered again and reviewed again, with 3 runs in all. If the third still fails you get the diagnostic, never a half-checked post.
Gate
What it checks
1. Capability
The reviewer agent and a way to get a hero image are there before writing starts
2. Format
.md, .html, .pdf and a valid hero exist, no leftover [IMAGE], [CHART] or [INTERNAL-LINK] markers, clean HTML
3. Visual
Screenshots at 375, 768 and 1280 pixels and in dark mode, valid JSON-LD, nothing overflowing
4. Review
blog-reviewer gives 90 or more with zero P0 issues, on the exact draft that ships
5. Links
Images resolve, links work, and the word count matches the schema
See a review: the scorecard the reviewer returns
## Quality Review: Best CRM for Freelancers
### Overall Score: 94/100 - Exceptional
| Category | Score | Max |
|-----------------------|-------|-----|
| Content Quality | 28 | 30 |
| SEO Optimization | 24 | 25 |
| E-E-A-T Signals | 14 | 15 |
| Technical Elements | 14 | 15 |
| AI Citation Readiness | 14 | 15 |
zero P0 issues found
Nonce: 3f9c1e...
BLOCKING: false (cleared review gate; 94/100 overall, no P0)
The nonce ties the review to the draft it read. Change the post and the gates ask for a new review. Example scores.
Yes. Claude Blog is MIT licensed and runs on your Claude Code plan. Gemini images, narration and NotebookLM are optional, and Claude Blog asks before using any of them.
Does it make up statistics?
It is built to stop that. Statistics must come from tier 1 to 3 sources with a date, the reviewer blocks unsupported claims that the post depends on, and /blog factcheck fetches each cited page to check the number is there. You should still read what you publish.
Do I need API keys?
No. Writing, reviewing and the gates work without keys, and the hero image falls back to free Openverse photos. Keys add options: Gemini for generated images and narration, Unsplash, Pexels or Pixabay for stock photos, and Google APIs for Search Console, PageSpeed and Analytics data.
Will my posts get cited by ChatGPT and Google AI Overviews?
Nobody can promise a citation. Claude Blog writes in the shape AI answers quote: a direct answer first, sourced numbers, clear headings, FAQ sections and schema, and the reviewer scores AI citation readiness out of 15. /blog geo checks any post you already have.
How is it different from asking Claude to write a post?
A prompt gives you one draft. Claude Blog runs a pipeline: research with sources, an outline, the draft, an independent review and five gates that code enforces, plus commands for everything around the post such as audits, refreshes, clusters and translations.
Does it work with WordPress, Next.js or Hugo?
Yes. It detects Next.js MDX, Astro, Hugo, Jekyll, WordPress, Ghost, 11ty, Gatsby and static HTML, and writes frontmatter, image paths and schema the way that platform expects.
Does my content leave my computer?
Claude Blog runs inside Claude Code, so your drafts are sent to Claude, the same as anything else you work on there. The researcher reads public web pages to find sources. Files are written to your machine. When you have no image keys, the free hero search sends the post title and a few tags to Openverse. No other service sees your content unless you turn on an optional API yourself.
Good to know: requirements, Blog Studio and a known warning
Requirements. Claude Code and Python 3.11 or newer. PDF rendering and screenshots need the presentation extra in pyproject.toml.
Blog Studio. An optional companion plugin in the same marketplace: a live status band, a pipeline and five-gate panel, a history of past posts and a Run checks button. Install it with /plugin install blog-studio@agricidaniel-blog. It lives on the blog-studio branch, needs claude-blog 2.3.2 or later, and the standalone installers never enable it.
Known warning.claude plugin validate on Claude Code 2.1.292 and later reports the name claude-blog as reserved. The plugin still installs and loads, and CI tolerates exactly that one finding. See docs/TROUBLESHOOTING.md.
What’s new
v2.3.2 makes Blog Studio work with the real engine, makes paid image calls opt-in in code and tightens the review gate. v2.3.1 added Blog Studio as an optional second plugin, and v2.3.0 brought the safe standalone installer, which backs up what it replaces and never deletes your own files. Full notes in the changelog.
MIT License, see LICENSE. Contributions are welcome: read CONTRIBUTING.md first and mention the checks you ran in your pull request. The animated scenes star the three bots from my GitHub profile; brand marks belong to their owners.
![claude-blog hero: in Claude Code you type /blog write with a topic, the bots search the web, tag tier 1 to 3 sources and throw a clickbait page in the trash, and a blog post writes itself line by line. The blog-reviewer scores it 87 with one P0 for an unsourced statistic, the green bot slaps on a source as citation [1], the score climbs to 94, all five delivery gates open and the post is delivered, ready for Google and AI answers.](https://api.gitread.org/api/repo-asset/AgriciDaniel/claude-blog/main/assets/readme/hero.svg?v=3)
评论