Writing

Overview

Writing defines the structured workflow for creating and editing content — articles, blog posts, documentation, research papers, social media posts, and project reports. It is loaded by the writer and editor agents and integrates with factcheck and humanize skills for quality assurance.

Content Pipeline

1. Research and Planning — define audience, purpose, key messages. Gather sources and data. Create a section-level outline. Identify claims that will need fact-checking.
2. Drafting — write from the outline, one section at a time. Lead with the most important information (inverted pyramid). Use concrete examples over abstract statements. Avoid AI writing patterns from the start.
3. Self-Review — read aloud mentally. Does every paragraph advance the argument? Is the structure scannable via headings? Are claims supported and cited?
4. Editorial Review — fact verification (extract and verify all claims), AI pattern scan (detect and remove artifacts), style consistency, structural review of argument flow.
5. Publication Preparation — final proofread, format for target platform, prepare multi-language versions, create backups.

Content Types and Registers

Each type has its own register: technical docs are formal and precise; blog posts are conversational and engaging (1000-3000 words); research papers are academic and rigorous (3000-10000 words); social media is casual and direct (100-500 words); legal documents are formal with numbered sections and defined terms; project reports are professional and structured (1000-5000 words). The skill adapts the pipeline to the target register.

Anti-Patterns

• Writing without an outline — leads to meandering structure
• Starting with "In this article..." — just start with content
• Symmetrical arguments without taking a position
• Filler conclusions ("The future looks bright")
• Citation-free claims — every non-obvious claim needs a source

Drafting Rules

• Numbers check — every number must come from the primary source (spec sheet, benchmark JSON), not from memory or context summaries
• Comparative metrics — if you write "37x faster", state what you are comparing against in the same sentence
• CTA platform check — before writing "tell us in the comments", verify the platform has comments
• Social posts must be self-contained — min ~200 words for FB/TG/LI. A picture does not replace text
• Verify links and pricing — check every URL is reachable before publishing; verify pricing and subscription claims by search, never from memory
• Every action gets a subject; scale the language to the event — name who did it, and keep routine operations in routine words
• No internal numbers that serve no reader; causal sanity check — include a figure only when it carries meaning, and test every stated cause against what the system actually does
• No announced headings, moralizing verdicts or template signposting — name the content, state the fact, translate jargon for the reader
• No pseudographic diagrams — no ASCII or box-drawing art as diagrams; complex figures get a real <figure> with informative alt text and a caption, equivalent in every language

Multi-Language Content

Write the primary version first, then translate with cultural adaptation, not literal translation. Each language version is independent and may have different structure, examples, or emphasis. Russian tech content uses English technical terms with Russian explanation. Long-form articles end on a specific CTA written in the version's own language.

Go-to-Market & Launch

The skill also carries launch checklists: technical SEO (indexing, canonical tags, titles and descriptions, JSON-LD, Core Web Vitals), analytics setup (GA4, Search Console, ad tracking), website pre-launch (HTTPS and HSTS, OG and Twitter Card tags, legal pages, PageSpeed), landing-page optimization and a campaign budget framework.