BlogFactory

Source-backed AI blog writer skill

Research-ready SEO drafts without invented evidence.

Source-Backed Blog Writer gives compatible AI clients a disciplined workflow for search-intent research, source review, claim mapping, article structure, metadata, and mechanical validation.

Install freeView on GitHub

CC BY 4.0 · Agent Skills compatible · Python standard-library validator

Source-Backed Blog Writer
Portable skill

Research before prose

Keep evidence attached to the draft.

The workflow stops when a material claim needs proprietary evidence instead of filling the gap with plausible-sounding copy.

01  inspect site + search intent
02  verify sources + map claims
03  draft + validate mechanically

Output: article, metadata, CMS excerpt, image recommendations, references, and a blocked, draft, or publish-ready label.

Why use it

Make the research contract visible before the article ships.

Sources before claims

Research the publishing site and current external sources, then map factual claims to numbered references.

Protect existing coverage

Check for competing pages and refresh useful material selectively instead of creating avoidable duplication.

Approval when needed

Use approval mode to review title options, target questions, claims, and the outline before drafting.

Validate the package

Run a dependency-free Python check for structure, metadata, citations, FAQs, images, and unresolved placeholders.

Workflow

One skill, four bounded stages.

  1. 01BriefTopic, domain, audience, market, voice, CTA, and real proprietary evidence.
  2. 02ResearchIntent, existing coverage, relevant pages, current sources, and content gap.
  3. 03DraftArticle pattern, complete draft, metadata, excerpt, images, and references.
  4. 04CheckMechanical validation plus a clear readiness state and remaining human review.

What is included

A complete writing workflow, not a CMS connection.

Research inputs

Topic, keyword, publishing domain, audience, market, language, voice, CTA, product detail, and supported firsthand evidence.

  • ✓Search-intent review
  • ✓Existing-page check
  • ✓Source and claim discipline

Article package

A structured 1,500–2,000-word target by default, with metadata and explicit readiness.

  • ✓CMS excerpt and slug
  • ✓Conditional contents and FAQs
  • ✓Images and numbered references

Client compatibility

The provider-neutral core follows Agent Skills conventions and keeps optional provider metadata separate.

  • ✓Codex and ChatGPT skill surfaces
  • ✓Claude Code, claude.ai, and API
  • ✓Other Agent Skills clients

Install free

Clone once, then copy the complete skill directory.

Keep SKILL.md, references/, and scripts/ together. Use a personal location for every project or a repository-local location for one workspace.

Open the repository

Codex installation
Terminal
git clone https://github.com/BlogFactoryHQ/source-backed-blog-writer-skill.git
mkdir -p ~/.agents/skills
cp -R source-backed-blog-writer-skill/source-backed-blog-writer ~/.agents/skills/

Claude Code: copy the same directory to ~/.claude/skills/. Other clients use their configured personal or project skills directory.

Questions

Know where the skill stops.

Is this a hosted AI writer?+

No. It is a portable set of Agent Skills instructions, references, and a validator. It uses the research and model capabilities available in your AI client.

Does it guarantee rankings or traffic?+

No. It improves source discipline and article completeness, but it cannot guarantee rankings, snippets, traffic, or inclusion in AI-generated answers.

Can it publish to a CMS?+

No. It produces an article package and readiness label. Use a separately authorized CMS workflow such as BlogFactory or Ghost Publisher only after review.

Which AI clients can use it?+

Its core follows Agent Skills conventions used by Codex, ChatGPT skill surfaces, Claude Code, claude.ai custom skills, the Claude API, and other compatible clients.

What happens when evidence is missing?+

The skill marks the package blocked or draft and names the missing evidence. It does not invent experience, quotes, data, credentials, or results.

What does the validator check?+

It checks mechanical requirements such as one H1, metadata, heading use, citations, FAQs, image fields, placeholders, and readiness format. Human source review is still required.

What is the license?+

Original repository content is openly licensed under CC BY 4.0. Third-party names, standards, websites, and documentation remain subject to their own terms.

Install the skill and start with a real brief.

Bring a topic, publishing domain, audience, CTA, and at least one verifiable proprietary fact.

Install free