Writer

Writer

Overview

Use this skill to produce and improve Google SEO-friendly articles that stay focused on one topic, use keywords naturally, and avoid invented facts or over-marketing. Default output language should match the user's request unless they specify another language.

When the requested destination is a Chinese WeChat Official Account (微信公众号、微信文章、公众号推文), load wechat-writer/SKILL.md and follow that workflow instead of applying this Google-first structure. When the destination is a LinkedIn Article, newsletter, LinkedIn long-form post, or LinkedIn thought-leadership package, use linkedin-writer/SKILL.md; its output contract overrides this file and stores the package under linkedin-writer/output/<article-slug>/. When the destination is Medium or another editorial third-party article, use medium-writer/SKILL.md; its output contract stores the package under medium-writer/output/<story-slug>/.

For a full reusable prompt and final article template, load references/seo-article-template.md when the user asks for a ready-to-copy prompt, a complete article, or a reusable writing template.

For article audits, existing-page reviews, or post-draft optimization, load references/seo-audit-checklist.md.

For fact-heavy articles, comparisons, product/category analysis, or style polishing, load references/fact-check-and-style.md.

After the SEO audit and audit-driven rewrite, load references/humanization.md for the final humanization pass. Preserve verified facts, search intent, required keywords, citations, code, URLs, and the author's voice while removing mechanical or overly AI-like writing patterns.

For deliverables that must be saved as files, use the output rules in this file and load references/output-packaging.md.

Cloudflare R2 is optional. Always create the article with local relative image paths first. If writer/config/r2.config.json exists and contains valid values, load references/r2-image-upload.md, upload the images, and replace the local paths with public URLs. If that local config is absent, keep the local image references and finish normally.

For Cloudflare R2 credential handling and leak-prevention rules, load references/r2-security.md.

Article Task Card

Before outlining or writing, create or infer a task card in 10 lines or fewer:

• Target reader: beginner, industry practitioner, or potential customer.
• Article type: how-to, comparison, listicle, or explainer.
• Main keyword: one primary keyword.
• Supporting keywords: 3 to 5 related keywords unless the user provides more.
• Target length: usually 800 to 1500 words.
• Tone: objective, editorial, restrained, and low-hype.
• Must include: at least 3 subheadings, 1 conclusion, and 3 FAQ items.
• Must avoid: fake data, fake quotes, unsupported claims, and excessive sales language.
• Optional CTA: link and anchor text if provided.
• Output format: Markdown unless the user asks otherwise.

If critical inputs are missing, make conservative assumptions and show them in the task card. Ask a question only when the topic, audience, or article type is too ambiguous to proceed safely.

Workflow

Follow this sequence for new articles:

1. Create the article task card.
2. Generate title options that naturally include the main keyword or a close grammatical variant, then select the clearest SEO title.
3. Generate the outline only: table of contents plus 1 to 3 bullet points for each section. Do not write body copy during this step unless the user explicitly asks for a full article immediately.
4. After the outline is approved or when the user asks to continue, write the article section by section in Markdown.
5. Generate article images with Codex's built-in image generation capability when available, then add the SEO meta pack: SEO Title, Excerpt, Meta Description, and Tags.
6. Run fact checks for objective claims, feature comparisons, statistics, definitions, dates, prices, legal or policy statements, and any claim that could mislead if wrong.
7. Run a writing SEO audit: check title, TDK, search intent, first 100 words, headings, keyword coverage, content depth, FAQ, internal-link opportunities, image alt suggestions, E-E-A-T signals, and unsupported claims.
8. Rewrite based on the audit, prioritizing factual accuracy, search intent, and high-impact SEO issues first.
9. Load references/humanization.md and humanize the audit-corrected draft. Match a user-provided voice sample when available; otherwise use a natural voice appropriate to the publication and audience.
10. Run a post-humanization integrity check on changed passages. Recheck claims, certainty, keyword placement, metadata lengths, citations, code, URLs, image paths, and CTA destinations.
11. Save the final Markdown article and image assets under writer/output/<article-slug>/.
12. Check for writer/config/r2.config.json. Upload only when it exists and is valid; otherwise keep local relative image paths without treating the missing config as an error.
13. Re-audit only the changed areas and report the main improvements, including the humanization pass.

When the user provides the topic and asks for a complete article in one go, still internally follow the outline-first structure, then output the full article without exposing unnecessary process notes.

Audit-Driven Optimization

Use the audit loop whenever a draft, published page, URL, or pasted article needs SEO improvement:

1. Separate deterministic checks from semantic judgment.
2. Deterministic checks include SEO title length, excerpt length, meta description length, H1 count, heading counts, keyword placement, FAQ presence, and missing image alt suggestions.
3. Semantic judgment includes search intent match, H1/title relevance, heading usefulness, content depth, E-E-A-T signals, clarity, specificity, and whether the CTA is too promotional.
4. Produce findings with this shape: issue, evidence, impact, fix, priority.
5. Rank fixes as Critical, High, Medium, or Low.
6. Apply fixes to the article instead of stopping at the report when the user asks to optimize.
7. Preserve true claims and the user's original positioning; do not fabricate facts to improve SEO.
8. Humanize the corrected draft only after Critical and High issues are resolved, then verify that the style rewrite did not change facts, certainty, links, code, or SEO requirements.

For URL audits, do not claim robots.txt, sitemap, PageSpeed, schema, canonical, status code, or crawlability results unless a tool or source actually verifies them. If technical checks are unavailable, mark them as "needs tool verification" and focus on content-level SEO.

Post-Audit Humanization

Humanization is the final editorial stage, not a substitute for fact-checking or SEO auditing. Load references/humanization.md after the audit-driven rewrite and follow its draft, critique, revision, and integrity-check sequence.

During this pass:

• Preserve the article's meaning, verified claims, limitations, citations, technical tokens, URLs, and required search intent.
• Use an approved writing sample for voice calibration when the user provides one. Match rhythm and register without copying distinctive phrases.
• Remove clusters of mechanical writing such as significance inflation, unsupported promotional language, vague attribution, repetitive transitions, forced three-item lists, synonym cycling, uniform paragraph rhythm, generic conclusions, chatbot artifacts, and excessive formatting.
• Prefer concrete details and direct verbs. Vary sentence and paragraph length when it improves reading, but do not introduce fake personality, deliberate errors, or theatrical fragments.
• Never invent first-person testing, personal experience, customer stories, quotes, opinions, or anecdotes to make the article sound human.
• Do not promise that the result is undetectable or can bypass AI detectors. The goal is reader quality and an appropriate editorial voice.

After rewriting, ask internally what still feels templated or over-produced, revise those passages once more, and recheck only the affected facts and SEO elements.

Fact Check and Evidence

Use a fact-check pass before finalizing articles that include claims beyond common knowledge:

1. Extract checkable claims: numbers, dates, rankings, definitions, product specs, policies, price, availability, legal/medical/financial statements, and competitive comparisons.
2. Mark each claim as verified, needs verification, softened, or removed.
3. Prefer primary or authoritative sources for verification: official documentation, product pages, standards bodies, government pages, research papers, or direct company statements.
4. Use web research when facts may have changed or when the user asks to reference current external analysis.
5. Synthesize external analysis in original language; do not copy another author's structure, phrasing, rankings, or conclusions directly.
6. Cite or name sources when factual support materially affects the reader's decision.
7. If reliable evidence is unavailable, rewrite the claim cautiously or omit it.

When using external sources, separate facts from interpretation. Facts should be traceable; recommendations can include the writer's own judgment as long as the reasoning is explicit.

When the user provides UI reference images to guide an article, use them as private reference material only. In the article body, captions, alt text, source note, SEO metadata, and audit notes, do not mention that visual references were provided or name their file/source format. Prefer natural phrasing such as "the interface shows," "the visible controls include," or "the live interface may change."

Objective Comparison

For comparison articles, make the comparison deep and evidence-aware:

• Define comparison criteria before judging options.
• Compare by user scenario, strengths, limitations, cost or effort, output quality, learning curve, integrations, risk, and best-fit use cases when relevant.
• Avoid declaring one option "best" without explaining for whom and under what conditions.
• Include trade-offs and limitations, not only benefits.
• Distinguish verified facts from editorial judgment.

Paragraph Quality Pattern

Use this paragraph pattern for main sections:

1. Conclusion sentence: state the practical point first.
2. Explanation: add 2 to 3 sentences with context or reasoning.
3. Example or steps: add 1 to 3 concrete examples, checks, or actions.
4. Mini-wrap: close with one sentence that connects back to the section topic.

Avoid filler openings, repeated definitions, and keyword stuffing. Use bullets only when they make scanning easier.

Type-Specific Rules

How-To

Use a four-step shape:

1. What to do: define the goal and destination.
2. What is needed: list materials, access, assumptions, or prerequisites.
3. How to do it: provide numbered steps.
4. Common questions: answer FAQ clearly.

Comparison

Compare across five dimensions:

• Who it is for.
• Ease of use.
• Output or result quality.
• Key features or practical differences.
• Recommendation by use case.

Listicle

Include 6 to 10 items unless the user asks for a different count. Each item should include a short explanation and a "best for" line when useful.

Explainer

Include a plain-language definition, why it matters, common misconceptions, and practical application.

SEO Requirements

• Include the main keyword or a natural close variant in the article title unless it would make the title grammatically awkward or misleading.
• Include the main keyword in the H1 or first paragraph.
• Include the main keyword in at least one H2.
• Include the main keyword once in the conclusion.
• Use each supporting keyword at least once naturally across the article.
• Let H2 headings be moderately longer when useful for SEO clarity, especially for comparison, tutorial, review, and news-analysis articles.
• Include the main keyword, a close variant, or a supporting keyword in H2 headings when natural; prioritize readability and search intent over exact-match stuffing.
• Generate an attractive SEO title related to the content, 65 to 70 characters.
• Generate a short excerpt related to the content, 155 to 160 characters.
• Generate an attractive SEO meta description related to the content, 55 to 160 characters.
• Prefer clear search intent over clever wording.
• Do not invent numbers, quotes, research, rankings, product specs, or citations.
• If a fact is uncertain, phrase cautiously with terms like "often," "in many cases," or omit it.

SEO Audit Requirements

Always evaluate these writing-level checks before final delivery:

• Search intent: title, intro, and sections answer the same query.
• Title/H1: clear, specific, and includes the main keyword naturally.
• First 100 words: main keyword and value proposition appear early.
• Heading structure: at least 3 useful subheadings; H2s are descriptive, specific, and may be longer when that helps include keywords naturally.
• Keyword distribution: main and supporting keywords appear naturally without repetition.
• Content depth: the article gives practical details, examples, steps, or comparison criteria.
• FAQ: at least 3 concise questions that match likely search follow-ups.
• SEO meta pack: SEO Title, Excerpt, and Meta Description are within limits, related to the content, attractive, and not generic.
• Image SEO: suggest relevant image types and alt text using natural keywords.
• Trust: add authorial clarity, limitations, prerequisites, or verification notes when relevant.
• Risk: flag unsupported data, vague claims, fake citations, and exaggerated marketing language.
• Humanization: after resolving audit findings, check voice consistency, sentence rhythm, repetitive templates, promotional tone, generic conclusions, and chatbot artifacts without changing verified meaning.