Claude Blog skill

Claude Blog writes and maintains blog posts from inside Claude Code, and it holds a draft back until the draft passes its own checks. It's a community project by Daniel Agrici under the MIT license, and the companion to his Claude SEO plugin. Anthropic doesn't publish it. If skills are new to you, start with the basics of Claude Skills.

What the skill does

One orchestrator, /blog, routes to 31 sub-skills and 5 agents in version 2.3.1. /blog write asks about audience, keyword and format, picks one of 12 templates and drafts with sourced statistics, images, charts and internal links. Its SKILL.md won't use the product-review template without documented first-hand testing, or the case-study one without measurable results and the subject's permission. It picks another template instead of faking the evidence. It also detects the blog platform, from Hugo and Astro to WordPress and plain HTML, and adjusts front matter to match.

Each draft is rendered to Markdown, HTML and PDF with a hero image, screenshotted at three screen widths, scored by a separate reviewer agent and checked for broken images and links. The delivery contract blocks any draft scoring under 90 out of 100 and retries up to three times before it hands the problem to you.

Beyond writing, four commands matter for a site that already has posts:

• /blog analyze scores an existing post from 0 to 100 and lists fixes by priority.
• /blog factcheck fetches the URLs a post cites and marks each claim as an exact match, a paraphrase or not found. Its SKILL.md says plainly that it checks whether a source says what the post claims, not whether the source is right.
• /blog decay compares two Search Console exports and flags pages whose clicks or impressions fell by 20% or more by default, with a hint to refresh, consolidate or prune.
• /blog cannibalization finds posts that overlap on keywords and suggests merging or separating them.

Other commands cover briefs, schema and translation with hreflang.

When to use it

• You publish often and want every post to go through the same checks before it's live.
• Old posts lost traffic and you have Search Console exports from two periods.
• A draft quotes numbers and each one should be traced back to its source.

For a quick outline in any chat, try our SEO blog outline prompt. To make a finished draft read less like machine output, run it through the Humanizer skill.

How to install it

In a Claude Code session, as a plugin:

/plugin marketplace add AgriciDaniel/claude-blog
/plugin install claude-blog@agricidaniel-blog

The README says the plugin was tested on Claude Code 2.1.296, and names 2.1.69 as the oldest version it can vouch for. Scoring and the gates need Python 3.11 or newer. There's also a standalone installer. The README asks you to pin it to a commit you've reviewed, and warns against the v2.2.0 tag, whose uninstaller deletes whole skill folders, including files you added. Restart Claude Code after installing.

Example request

/blog write [topic]
Audience: [who reads it]. Primary keyword: [keyword]. Format: [Markdown / MDX / HTML]. Cite only sources you can link. If a template needs first-hand testing I haven't done, pick another one and tell me which you chose.

Limits

• The maintainer's own note in the README says direct prompting is faster for a single throwaway draft, and that hosted writing tools are easier for non-developers.
• Its AI citation score is, by the README's own label, non-calibrated: a writing heuristic that doesn't measure whether AI tools cite you.
• Hero images come from a chain of sources, starting with a Banana MCP server and the Gemini API and ending at stock APIs and Openverse. Audio narration makes billed Gemini calls.
• The README reports that Claude Code 2.1.292 and later flag the name claude-blog as reserved during claude plugin validate. It still installed in the maintainer's tests, but a later release could refuse it.
• Drafts include a "Key Takeaways" box by default, so check the output against your own style rules before publishing.