How to use the Blog Draft Writer agent skill (blog-draft)

What the blog-draft skill makes the agent do

When you ask the agent to write a blog post, the skill tells it to work in stages instead of producing a finished article at once. It creates a dated project folder, gathers and summarises your sources (links, files, or web searches), and proposes angles and key points. It also asks clarifying questions about your goal, target length and what to cut.

After you approve the outline, the agent writes a full draft with inline citations and a references list. The draft is saved as draft-v0.1.md. When you ask for changes, the agent saves each revision as a new numbered file so you can compare versions.

• Creates a folder named YYYY-MM-DD plus a short topic slug, with a resources/ subfolder
• Writes one summary file per source in resources/
• Saves the approved outline as OUTLINE.md
• Saves drafts as draft-v0.1.md, draft-v0.2.md and so on
• Asks for your approval at each stage

Install the blog-draft skill

The skill ships as a ZIP download from the directory page. Choose one of the two install routes below. Skills are loaded from a folder that contains a SKILL.md file, so keep the unzipped folder intact.

For a personal install, unzip the download into ~/.claude/skills/ so the skill is available in every project. For a single project, unzip it into that project's .claude/skills/ folder instead.

If you use the Claude apps, you can upload the ZIP under Settings → Capabilities → Skills rather than unzipping it yourself.

• Personal install: unzip into ~/.claude/skills/
• Project install: unzip into .claude/skills/ inside your project
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Trigger the skill with a natural request

You do not need a special command. The skill's description tells the agent to use it when you want to write a blog post, create content from research, or draft an article. Describe your topic and, if you have them, add your sources, audience and tone.

Good requests name the topic and give the agent material to work from. You can also say how long the post should be. The skill's length guide is short (500-800 words), medium (1000-1500 words) or long (2000+ words).

• "Write a blog post about remote onboarding. Use these three links as sources."
• "Help me draft an article for junior developers on code review. Keep it friendly and around 1000 words."
• "Update the existing post in blog-posts/ with the feedback below."

Work through the approval steps

The skill is built around checkpoints. The agent pauses after the research summary, after the outline, and after each draft, and waits for your reply before continuing. If you skip a checkpoint, the agent is designed to wait rather than push ahead.

To update a post that already exists, tell the agent so. The skill then skips the setup steps and starts at the revision step, reading the existing draft files first.

• After research: review the source summaries and answer the clarifying questions
• After the outline: approve it, or ask for edits before it is saved
• After each draft: give feedback, or accept the draft to finish
• On acceptance: the agent can rename the final draft to final.md if you ask

What the skill needs and what it does on your machine

The skill needs web search to research a topic and network access to read URLs you provide. Git is optional. If the blog folder is inside a git repository, the agent commits the outline and drafts. If it is not, the agent skips those commits and tells you.

The skill's files include only SKILL.md. It does not run scripts. Pushing commits requires access to a git remote, and the skill only pushes if the agent is set up to do so.

• Web search capability for topic research
• Network access to fetch the URLs you give it
• Git, optional: used only if the folder is a git repo
• A git remote, only if you want commits pushed

Tips for better drafts and safety

The skill asks for a citation on every statistic, comparison and factual claim, and the draft should list its sources at the end. Citations make checking easier, but the skill does not verify that a source is accurate. Open the sources and confirm the key claims before you publish.

Give the agent good sources and state your goal clearly. Keep feedback specific, since each revision is saved as a new version and you can go back to earlier files.

• Check every cited figure against the original source before publishing
• Provide your own links and files rather than relying only on web search
• Name the reader and tone in your first request
• Give feedback in one message per round so revisions stay easy to compare