blog-post-writing

name: blog-post-writing description: "Write and iteratively refine Japanese-language technical blog posts — wiki research first, humanizer review, multi-round user feedback integration." version: 1.0.0 metadata: hermes: tags: [blog, writing, japanese, humanizer, wiki-research, creative] category: creative related_skills: [humanizer, raw-article-filename-policy, llm-wiki]

Blog Post Writing

Write and iteratively refine Japanese technical blog posts using wiki knowledge as the primary source, Humanizer for review, and multi-round user feedback for structural direction.

When to Use

Load this skill when the user:

• Asks to write a blog post, essay, or article on a technical AI topic
• Requests a "活き活きとした" (lively) or "深掘り" (deep-dive) blog post
• Provides multi-round structural feedback ("reduce emphasis on X," "fold Y into Z," "add section on W")
• Wants Japanese output

Core Workflow

Phase 1: Research

1. Wiki first. Search ~/ai-topics/wiki/ for relevant concepts, entities, comparisons. Prioritize wiki knowledge over web search. The wiki has deep, cross-referenced content that produces richer arguments than surface-level web results.

2. Web fallback. If the wiki lacks support for a specific claim, use web_search and web_extract to find authoritative sources. Cite them.

3. Build the argument structure. Before writing, identify:

   • The core thesis (one sentence)
   • 3-5 supporting arguments drawn from wiki concepts
   • The narrative arc (chronological? layered? compare-contrast?)

Phase 2: Writing

4. Write to blog/. Save files to ~/ai-topics/blog/{YYYY-MM-DD}_{author}_{short-slug}.md. Do NOT save to wiki/raw/articles/ — that directory is for externally-sourced article scrapes only.

5. Frontmatter. Include title, date, author, tags, sources (wiki pages + external URLs), and optional series / series_index if part of a multi-post sequence.

6. Japanese voice. Write in natural Japanese. Mix registers (である調 + だ調 + occasional です・ます for reader address). Vary paragraph length. Avoid the "AI triple combo" — —— in every header + **bold** everywhere + rigid である調.

Phase 3: Humanizer Review

7. Always run through humanizer. After writing, load the humanizer skill and apply BOTH the 29 English patterns AND the Japanese-specific patterns (J1-J8). The humanizer skill already contains a comprehensive Japanese section.

8. Two-pass review:

   • Pass 1: Draft rewrite with all pattern fixes
   • Pass 2: "What makes this so obviously AI generated?" → fix remaining tells

9. Save the humanized version as a separate file or overwrite, depending on user preference. Default: save as separate -humanized.md variant for comparison.

Phase 4: User Feedback Integration

10. Multi-round structural feedback is normal. The user will often give layered feedback:

    • "Reduce emphasis on X, it's too prominent"
    • "Fold section Y into Z, it's too detailed as standalone"
    • "Add a new angle on W"
    • "Remove section Q entirely, it doesn't fit the thesis"

11. Treat each round as structural editing, not full rewrite. Preserve the voice you've established. The user is steering content and emphasis, not rejecting tone. Apply targeted patches rather than rewriting from scratch.

12. When the user questions a logical leap ("主張に飛躍がある"): verify the wiki source more carefully, identify the missing logical steps, and make the chain explicit. This is a content quality signal, not a voice signal.

13. When the user proposes a new angle without deep background ("深い考えがあるわけではないので、まず考察を行なって"): research the angle using wiki + web, present the analysis to the user for validation, then weave it into the post.

File Management

Directory

~/ai-topics/blog/
  2026-05-08_hermes_tradeoff-v1.md         # initial draft
  2026-05-08_hermes_tradeoff-humanized.md   # after humanizer pass
  2026-05-08_hermes_tradeoff-v2.md          # after structural revision
  2026-05-08_hermes_tradeoff-v3.md          # after further revision
  2026-05-08_hermes_prediction-vs-valuation.md  # follow-up post

Naming Convention

{YYYY-MM-DD}_{author}_{short-slug}.md

• author = hermes for agent-authored posts
• short-slug = 2-4 word kebab-case summary of the topic
• Versioned files append -v2, -v3, etc. to the slug

Commit

After writing or editing, always commit + push:

cd ~/ai-topics && git add blog/ && git commit -m "blog: <summary>" && git push

Series Posts

When writing a multi-post series:

• Add series: <series-slug> and series_index: <N> to frontmatter
• Cross-reference earlier posts with relative links
• The first post establishes the thesis; follow-ups explore orthogonal dimensions or deeper dives

Pitfalls

• Don't save to wiki/raw/articles/. That directory is for external article scrapes. Blog posts go to blog/.
• Don't over-use the humanizer. The goal is natural writing, not sterile anti-AI writing. Japanese technical blogs are structurally more organized than English blogs — numbered sections and clear transitions are normal and should be preserved.
• Don't silently rewrite on user feedback. Show what changed. Use targeted patch when possible; use write_file for full rewrites only when the structure fundamentally changes.
• Don't treat every user comment as a full-structure rejection. "Reduce emphasis on X" means reduce, not delete. "Fold Y into Z" means merge, not rewrite both.
• Verify wiki claims before citing them. If a wiki concept page has gaps or errors, note them to the user rather than propagating them into the blog.