blog-from-chat

name	blog-from-chat
description	Write a polished blog post from a chat or conversation summary, with multi-language output. Use when the user wants to turn a Claude conversation, chat log, or discussion into a publishable blog article, or mentions "blog from this chat" or "write up our conversation".

Blog from Chat Summary (Generic)

This skill generates a blog post from a chat summary. It supports Turkish, English, or both languages.

Generic Design: Works for any user with GitHub auth. Reads configuration from memory.

Input

Chat summary provided as $ARGUMENTS.

Configuration (Read from Config & Memory)

Before starting, read ~/.claude/config.json for:

• BLOG_DIR - Blog directory path (from paths.blog_dir)
• USERNAME - GitHub username (from user.username)
• DEFAULT_LANGUAGE - Default language setting (from blog.default_language)

Fallback detection:

• If config doesn't exist: Use mcp__github__get_me()["login"] and ask user for blog_dir

Workflow

Step 1: Ask Language Preference (FIRST)

Use AskUserQuestion to ask:

Blog gönderisi hangi dilde olsun?

1. Türkçe
2. İngilizce
3. Hem Türkçe hem İngilizce (iki ayrı post)

Default: Option 1 (Turkish) if not specified.

Step 2: Learn Writing Style

Read the last 2-3 blog posts from $BLOG_DIR/_posts/ to understand:

• Tone and voice (friendly, professional, first-person)
• Structure (Problem → Solution → Implementation → Results)
• Formatting patterns
• Image placement (~1 image per 50 lines)

Step 3: Generate Frontmatter

Based on the chat summary, create appropriate frontmatter:

---
title: "Generated Title"
description: "Brief 1-2 sentence description"
date: YYYY-MM-DD HH:MM:SS +0300
categories: [Category1, Category2]
tags: [tag1, tag2, tag3, tag4]
image:
  path: /assets/img/YYYY-MM-DD-slug/banner.png
  alt: "Banner image description"
---

Step 4: Estimate Line Count

Calculate and show estimate:

# Estimate based on chat summary length
estimated_lines = len(chat_summary.split()) / 3  # Rough estimate

print(f"Tahmini blog uzunluğu: ~{estimated_lines} satır")
print("Onaylıyor musunuz? Veya tercih ettiğiniz aralığı belirtin (örn: 200-250 satır)")

Wait for user confirmation or preferred range.

Step 5: Write Blog Content

Generate blog content in the selected language following the learned style.

Adjust length based on user's preferred range if specified.

Step 6: Find and Download Images

Use your configured image-search MCP (e.g. a Google / Bing image-search MCP server):

1. Search for banner image (main topic related)
2. Search for inline images (~1 per 50 lines of content)
3. Download all images to $BLOG_DIR/assets/img/YYYY-MM-DD-slug/
4. Use format: banner.png, image1.png, image2.png, etc.

Skip this step if no image-search MCP is configured.

Step 7: Show Draft

Display the complete draft in the terminal using fenced code blocks.

Step 8: Wait for User Approval

Ask user: "Taslak onaylanıyor mu? Blog dosyası oluşturulsun mu?"

Step 9: Write Blog File (On Approval)

If approved:

1. Create the file at $BLOG_DIR/_posts/YYYY-MM-DD-slug.md
2. If bilingual: Create two files (-tr.md and -en.md)

Step 10: Create DEV.to Draft (Optional)

Skip this step entirely if no DEV.to publishing MCP is configured. Otherwise, use it to create the article draft:

• Set published: false
• Use appropriate tags (max 4)
• Include frontmatter as description

Step 11: Ask for DEV.to Publish Approval

Ask user: "DEV.to'da publish edilsin mi?"

Step 12: Publish to DEV.to (On Approval)

Update the draft to published: true via the same MCP used in Step 10.

Critical Requirements

• Generic Design: No hardcoded usernames or paths
• Read config from config.json: User settings in ~/.claude/config.json
• Auto-detect username: Use config["user"]["username"] or fallback to mcp__github__get_me()["login"]
• Language selection is FIRST question - always ask before generating content
• Read previous blog posts to match writing style
• Show draft first - never auto-write files
• Separate approvals for file write and DEV.to publish
• Default to Turkish unless user specifies otherwise
• Download images to correct directory structure if an image-search MCP is configured

Image Guidelines

• Banner: 1200x630px PNG
• Inline images: Every ~50 lines
• Save to: $BLOG_DIR/assets/img/YYYY-MM-DD-slug/
• Always include alt text

Memory Reference

Config Path: ~/.claude/config.json This file contains user settings (username, blog directory, language).

Variables Used

# Read from config or auto-detect
from lib.config import get_config
config = get_config()
USERNAME = config["user"]["username"] or mcp__github__get_me()["login"]
BLOG_DIR = config["paths"]["blog_dir"]

# Derived paths
POSTS_DIR = f"{BLOG_DIR}/_posts/"
IMAGES_DIR = f"{BLOG_DIR}/assets/img/"