Blog Post Generator

Set up a GitHub Actions workflow that uses Claude Code to automatically write blog posts and open PRs.

Trigger

Use when user says "set up blog automation", "auto blog generator", "blog post github action", "automated content pipeline", "auto-generate blog posts", or asks about automating blog content creation with CI/CD.

Goal

Create a complete GitHub Actions workflow that:

1. Runs on schedule (cron) or manual dispatch
2. Uses anthropics/claude-code-base-action@beta to generate a blog post
3. Optionally calls Nano Banana (gemini-2.5-flash-image) for hero image generation
4. Commits content to a new branch and opens a PR for review

Process

Step 1: Gather Requirements

Ask the user about their blog setup. Do not proceed until you have answers for at least items 1-4:

1. Blog content directory — where posts live (e.g., content/blog/, src/posts/, app/blog/)
2. Frontmatter format — what fields are required:
   • title, date, slug, description, image, tags, author, category, draft
   • Are there custom fields specific to their framework?
3. Content framework — what renders the blog:
   • Next.js MDX (*.mdx files)
   • Astro (.md or .mdx)
   • Hugo (.md with TOML/YAML frontmatter)
   • Jekyll (.md with YAML frontmatter)
   • Plain markdown
4. Tone and style — writing guidelines:
   • Formal / casual / technical / conversational
   • Target audience (developers, marketers, founders, general)
   • Typical post length (500-1000, 1000-2000, 2000+ words)
   • Any forbidden patterns (e.g., no fluff, no "in today's world")
5. Hero image generation — do they want Nano Banana image gen?
   • Requires GEMINI_API_KEY secret
   • Generates 16:9 PNG images for each post
6. Schedule — how often:
   • Weekly (e.g., every Monday at 9am UTC)
   • Daily
   • Manual only (workflow_dispatch)
   • Custom cron expression
7. Topic source — where topics come from:
   • Manual input via workflow dispatch
   • Topic queue file (content/topics.yml)
   • Keyword list that Claude picks from

Step 2: Create the Prompt File

Create .github/prompts/blog-post.md containing the blog writing instructions for Claude.

This file is passed to claude-code-base-action via the prompt_file parameter and controls how Claude writes each post.

# Blog Post Generator Prompt

You are writing a blog post for [PRODUCT/COMPANY].

## Audience
[Target audience description from Step 1]

## Tone & Style
- [Tone from Step 1: e.g., "Technical but accessible, conversational"]
- Write in active voice
- Use short paragraphs (2-3 sentences max)
- Include concrete examples and data where possible
- No filler phrases: never use "in today's world", "it's no secret that", "let's dive in", "without further ado"
- No generic intros — start with a hook or a specific claim

## Structure
1. **Title** — compelling, keyword-aware, under 60 characters
2. **Meta description** — action-oriented, under 155 characters
3. **Introduction** (2-3 paragraphs) — state the problem or insight, why it matters now
4. **Body sections** (3-5 H2 sections) — each section covers one key point with:
   - A clear H2 heading
   - Supporting evidence, examples, or code snippets
   - Subheadings (H3) where needed for scannability
5. **Conclusion** — summarize key takeaway, include a soft CTA
6. **Frontmatter** — must include all required fields

## Frontmatter Template
```yaml
---
title: "[Post Title]"
date: "[YYYY-MM-DD]"
slug: "[url-safe-slug]"
description: "[Meta description under 155 chars]"
image: "/blog/[slug]/hero.png"
tags: ["tag1", "tag2"]
author: "[default author]"
draft: false
---

File Naming

Save the post as: [CONTENT_DIR]/[YYYY-MM-DD]-[slug].mdx

SEO Guidelines

• Use the target keyword in the title, first paragraph, and at least one H2
• Include 2-3 internal links to other pages on the site
• Add 1-2 external links to authoritative sources
• Use descriptive alt text for any images referenced
• Keep the slug short and keyword-focused

What to Read First

Before writing, read:

• CLAUDE.md for project-specific writing rules
• Recent posts in the content directory for style reference
• The topic brief (provided as input or from content/topics.yml)

Adapt this template based on the user's answers from Step 1. Replace all `[PLACEHOLDER]` values with their actual configuration.

### Step 3: Create the Hero Image Script (Optional)

If the user wants hero image generation, create `.github/scripts/generate-hero-image.mjs`:

```javascript
#!/usr/bin/env node

/**
 * Generate a hero image for a blog post using Nano Banana (Gemini image generation).
 *
 * Usage: node generate-hero-image.mjs <blog-post-path>
 *
 * Requires GEMINI_API_KEY environment variable.
 * Falls back gracefully if the key is not set.
 */

import { GoogleGenAI } from "@google/genai";
import { readFileSync, writeFileSync, mkdirSync } from "node:fs";
import { dirname, join, basename } from "node:path";

const postPath = process.argv[2];
if (!postPath) {
  console.error("Usage: node generate-hero-image.mjs <blog-post-path>");
  process.exit(1);
}

const apiKey = process.env.GEMINI_API_KEY;
if (!apiKey) {
  console.log("GEMINI_API_KEY not set — skipping hero image generation.");
  process.exit(0);
}

// Extract frontmatter from the blog post
const content = readFileSync(postPath, "utf-8");
const frontmatterMatch = content.match(/^---\n([\s\S]*?)\n---/);
if (!frontmatterMatch) {
  console.error("No frontmatter found in", postPath);
  process.exit(1);
}

const frontmatter = frontmatterMatch[1];
const title = frontmatter.match(/title:\s*"?([^"\n]+)"?/)?.[1] || "Blog Post";
const description =
  frontmatter.match(/description:\s*"?([^"\n]+)"?/)?.[1] || title;

// Generate the image
const ai = new GoogleGenAI({ apiKey });

const imagePrompt = [
  `Create a professional, modern blog hero image for an article titled "${title}".`,
  `The article is about: ${description}.`,
  "Style: clean, minimal, abstract or conceptual illustration.",
  "Use a professional color palette with good contrast.",
  "Do NOT include any text, words, letters, or typography in the image.",
  "Aspect ratio: 16:9, suitable for a blog header.",
  "The image should feel premium and editorial, like a top-tier tech blog.",
].join(" ");

console.log("Generating hero image for:", title);

const response = await ai.models.generateContent({
  model: "gemini-2.5-flash-preview-05-20",
  contents: [{ role: "user", parts: [{ text: imagePrompt }] }],
  config: {
    responseModalities: ["TEXT", "IMAGE"],
  },
});

// Find the image part in the response
const imagePart = response.candidates?.[0]?.content?.parts?.find(
  (p) => p.inlineData?.mimeType?.startsWith("image/")
);

if (!imagePart) {
  console.error("No image generated in response.");
  process.exit(1);
}

// Determine output path from frontmatter image field or derive from post path
const slug = basename(postPath, ".mdx")
  .replace(/^\d{4}-\d{2}-\d{2}-/, "");
const imageDir = join(dirname(postPath), "..", "public", "blog", slug);
mkdirSync(imageDir, { recursive: true });
const imagePath = join(imageDir, "hero.png");

// Decode and save the image
const imageBuffer = Buffer.from(imagePart.inlineData.data, "base64");
writeFileSync(imagePath, imageBuffer);
console.log("Hero image saved to:", imagePath);

// Update frontmatter with the image path
const relativeImagePath = `/blog/${slug}/hero.png`;
const updatedContent = content.replace(
  /image:\s*"[^"]*"/,
  `image: "${relativeImagePath}"`
);
writeFileSync(postPath, updatedContent);
console.log("Updated frontmatter image path to:", relativeImagePath);

The script paths (image output directory, frontmatter image field) should be adapted to match the user's project structure from Step 1. The example above assumes a Next.js-style public/ directory.

Step 4: Create the GitHub Action Workflow

Create .github/workflows/blog-generator.yml:

name: Blog Post Generator

on:
  schedule:
    # Runs weekly on Monday at 9:00 UTC — adjust to user preference
    - cron: "0 9 * * 1"
  workflow_dispatch:
    inputs:
      topic:
        description: "Blog post topic or title"
        required: false
        type: string
      generate_image:
        description: "Generate hero image with Nano Banana"
        required: false
        type: boolean
        default: true

permissions:
  contents: write
  pull-requests: write

concurrency:
  group: blog-generator
  cancel-in-progress: false

jobs:
  generate-post:
    runs-on: ubuntu-latest
    timeout-minutes: 30
    steps:
      - name: Checkout repository
        uses: actions/checkout@v4
        with:
          fetch-depth: 0

      - name: Generate blog post with Claude
        id: claude
        uses: anthropics/claude-code-base-action@beta
        with:
          prompt: |
            Write a new blog post.
            ${{ github.event.inputs.topic && format('Topic: {0}', github.event.inputs.topic) || 'Pick the next unwritten topic from content/topics.yml. If no topics file exists, write about a relevant topic for the product.' }}

            Follow the instructions in .github/prompts/blog-post.md exactly.
            Read existing posts for style reference before writing.
            Create the post file in the correct directory with proper frontmatter.
          prompt_file: ".github/prompts/blog-post.md"
          model: "claude-sonnet-4-6"
          max_turns: 15
          allowed_tools: "Read,Write,Edit,Bash(git:*),Bash(ls:*),Bash(date:*),Glob,Grep"
        env:
          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}

      - name: Find new blog post
        id: find-post
        run: |
          # Find the most recently created .mdx or .md file in the content directory
          NEW_POST=$(git diff --name-only --diff-filter=A HEAD | grep -E '\.(mdx?|md)$' | head -1)
          echo "post_path=$NEW_POST" >> "$GITHUB_OUTPUT"
          echo "Found new post: $NEW_POST"

      - name: Generate hero image (Nano Banana)
        if: |
          (github.event.inputs.generate_image == 'true' || github.event_name == 'schedule') &&
          steps.find-post.outputs.post_path != ''
        run: |
          npm install @google/genai
          node .github/scripts/generate-hero-image.mjs "${{ steps.find-post.outputs.post_path }}"
        env:
          GEMINI_API_KEY: ${{ secrets.GEMINI_API_KEY }}

      - name: Create Pull Request
        if: steps.find-post.outputs.post_path != ''
        env:
          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
        run: |
          # Extract slug from the post filename
          POST_FILE="${{ steps.find-post.outputs.post_path }}"
          SLUG=$(basename "$POST_FILE" | sed 's/\.[^.]*$//' | sed 's/^[0-9]\{4\}-[0-9]\{2\}-[0-9]\{2\}-//')
          DATE=$(date +%Y-%m-%d)
          BRANCH="blog/${DATE}-${SLUG}"

          git checkout -b "$BRANCH"
          git add -A
          git commit -m "blog: add post — ${SLUG}"
          git push origin "$BRANCH"

          gh pr create \
            --title "Blog: ${SLUG}" \
            --body "$(cat <<'EOF'
          ## New Blog Post

          Auto-generated blog post via Claude Code.

          **Post**: \`${{ steps.find-post.outputs.post_path }}\`

          ### Checklist
          - [ ] Review content for accuracy
          - [ ] Check frontmatter fields
          - [ ] Verify internal/external links
          - [ ] Review hero image (if generated)
          - [ ] Preview in local dev environment

          *Generated by [blog-generator](https://github.com/vood/growth-marketing) workflow*
          EOF
          )" \
            --base main \
            --head "$BRANCH"

Adapt the workflow based on Step 1 answers:

• Schedule: replace the cron expression with the user's preference
• Content directory: adjust the grep pattern in the find-post step
• Model: claude-sonnet-4-6 for cost efficiency, or claude-opus-4-6 for higher quality
• Hero image step: remove entirely if user doesn't want image generation
• Topic source: modify the prompt to match user's topic strategy

Step 5: Create CLAUDE.md Additions

Add blog-specific instructions to the project's CLAUDE.md (or create one if it doesn't exist). These instructions guide Claude when running in CI:

## Blog Content Guidelines

When generating blog posts in CI:

### Writing Rules
- Write for [TARGET AUDIENCE] — assume they are [EXPERTISE LEVEL]
- Tone: [TONE FROM STEP 1]
- Length: [WORD COUNT RANGE] words
- Every claim needs evidence: a link, a stat, or a concrete example
- No filler paragraphs — every section must deliver value

### Forbidden Patterns
- "In today's fast-paced world..."
- "It's no secret that..."
- "Let's dive in / dive deep"
- "Without further ado"
- "In conclusion" (just conclude, don't announce it)
- "Game-changer", "revolutionize", "cutting-edge" (unless literally true)
- Starting paragraphs with "So," or "Now,"

### Required Frontmatter
Every post must include:
- title (under 60 chars)
- date (YYYY-MM-DD)
- slug (URL-safe, keyword-focused)
- description (under 155 chars, for meta description)
- image (path to hero image)
- tags (2-5 relevant tags)
- author

### Image Conventions
- Hero images: `/blog/[slug]/hero.png`
- Inline images: `/blog/[slug]/[descriptive-name].png`
- Always include alt text

### Internal Linking
- Link to at least 2 other pages on the site
- Link to relevant tool pages under `/tools/` when applicable
- Use descriptive anchor text (not "click here")

Step 6: Research Competitors for Topic Ideas

Before populating the topic queue, research competitor and adjacent company blogs to find what topics rank well, identify content gaps, and generate high-value topic ideas.

6a: Identify Competitors & Adjacent Companies

Ask the user for their competitors, or discover them:

1. Direct competitors — tools in the same space as the user's product
2. Adjacent tools — related categories that share the same target audience
3. Industry thought leaders — blogs that cover the same topics

For each, look for their /blog, /resources, /learn, or /guides sections.

6b: Catalog Competitor Blog Content

For each competitor blog, search the web and catalog:

Company	Blog URL	Top Topics	Reddit/AI-Related Posts	Estimated Frequency

Focus on:

• Their most-linked and most-shared posts (indicates what resonates)
• Posts targeting keywords relevant to the user's product
• Content formats they use (guides, comparisons, how-tos, data studies, listicles)
• Topics they cover repeatedly (signals proven demand)

6c: Find Content Gaps

Identify topics that:

• No competitor covers well — thin content, outdated posts, or missing entirely
• Have search demand — check Google autocomplete, "People Also Ask", and related searches for queries like:
  • "[product category] + guide/tips/strategy/tools/examples"
  • "[competitor name] + alternatives/vs/review"
  • "[target audience] + [problem the product solves]"
  • "how to [action the product enables]"
• Are trending — new developments, emerging trends, or recent changes in the industry that competitors haven't covered yet

6d: Generate & Prioritize Topic Ideas

For each topic idea, provide:

### [Topic Title]
- **Target keyword**: the primary search query this would rank for
- **Search intent**: informational / commercial / comparison
- **Content format**: how-to guide / listicle / comparison / data study / opinion
- **Brief**: 2-3 sentences on what the post should cover
- **Why this matters**: competitive gap, search demand signal, product relevance
- **Competitors covering this**: who has similar content (or "none")
- **Differentiation**: what angle or insight makes our version better

Rank all ideas by:

1. SEO potential — search volume indicators and ranking difficulty
2. Product relevance — how naturally the topic leads to the user's product
3. Content gap — how poorly competitors cover it (bigger gap = bigger opportunity)
4. Freshness — timely topics that capitalize on recent trends

Present the top 15-20 topics as the recommended topic queue.

6e: Save Research Output

Save the full research findings to research-output/blog-topic-research.md with:

1. Competitor blog inventory (what they publish, how often, top topics)
2. Content gap analysis
3. Keyword opportunities found
4. Prioritized topic list with full details

Step 7: Create Topic Queue File

Create content/topics.yml populated with the prioritized topics from Step 6:

# Blog Topic Queue
# Claude picks the next topic with status: pending
# After writing, Claude updates status to: written

topics:
  - title: "How to Monitor Reddit Mentions for Your Brand"
    keyword: "reddit brand monitoring"
    intent: informational
    format: how-to guide
    brief: "Guide for marketers on setting up Reddit mention tracking. Cover manual methods vs automated tools. Include real examples of brands responding to Reddit mentions."
    status: pending

  - title: "Share of Voice on Reddit: What It Is and How to Measure It"
    keyword: "reddit share of voice"
    intent: informational
    format: data study
    brief: "Explain SOV in the context of Reddit discussions. How to calculate it, why it matters for competitive intelligence, and how to improve it."
    status: pending

  - title: "Reddit Marketing Strategy for B2B SaaS Companies"
    keyword: "reddit marketing b2b saas"
    intent: commercial
    format: guide
    brief: "Tactical guide for B2B SaaS companies. Which subreddits to target, how to engage authentically, measuring ROI. Include dos and don'ts."
    status: pending

Each topic should have:

• title — working title (Claude may refine it)
• keyword — primary SEO keyword to target
• intent — informational, commercial, or comparison
• format — how-to guide, listicle, comparison, data study, opinion
• brief — 2-3 sentences describing what the post should cover
• status — pending (unwritten), written (generated), published (live)

Required Secrets

Remind the user to add these GitHub repository secrets:

Secret	Required	Purpose
ANTHROPIC_API_KEY	Yes	Powers Claude Code for blog writing
GEMINI_API_KEY	No	Powers Nano Banana for hero images

GitHub's GITHUB_TOKEN is automatically available and handles PR creation.

Verification

After creating all files, verify:

1. .github/workflows/blog-generator.yml — valid YAML, correct action references
2. .github/prompts/blog-post.md — all placeholders replaced with user's config
3. .github/scripts/generate-hero-image.mjs — correct paths for user's project structure
4. CLAUDE.md — blog guidelines added (not overwriting existing content)
5. content/topics.yml — valid YAML with relevant topics (if using topic queue)

Run the workflow manually via workflow_dispatch to test before relying on the schedule.

Tips

• Start with workflow_dispatch only (no schedule) until you've reviewed a few generated posts
• Use claude-sonnet-4-6 for cost-effective generation; switch to claude-opus-4-6 for flagship content
• Keep the prompt file detailed — the more specific your instructions, the better the output
• Review and merge generated PRs promptly to keep the topic queue moving
• Add a max_turns: 15 to prevent runaway generation; increase if posts are being cut short
• The hero image script uses gemini-2.5-flash-preview-05-20 — update the model ID as newer versions release
• Consider adding a linting step (markdownlint, frontmatter validator) before PR creation