write-content

这段内容是审核时保存的快照。外部来源才是完整且最新的版本。

---
name: write-content
description: Use when writing a complete SEO article. Includes the full anti-AI-slop ruleset (banned vocabulary, banned phrases, banned structural patterns) and voice rules. The agent researches the SERP itself if needed — no keyword data exports required.
---

# Write Content

Writes a complete SEO-optimized article. Four phases: research → content type decision → knowledge extraction → write. Includes the full anti-AI-slop ruleset and the voice rules that make the output sound like a practitioner, not a press release.

## Input

- **Topic or target keyword** (required)
- *(Optional)* An existing content brief — skip the research phase if provided
- *(Optional)* Expert interview output from the `expert-interview` skill

If no topic is given, ask for one before proceeding.

## Business context persistence

Business context (audience, tone, language, brand voice, examples) shapes every article. Don't re-ask these questions every session.

**First use**: ask 4-5 questions and save the answers somewhere persistent in your agent's environment. Use the safest default first:

- **Claude Code**: `~/.claude/projects/<path>/memory/business-context.md` is the recommended default. Do NOT write to `./CLAUDE.md` unless the user explicitly asks for it — `CLAUDE.md` is the user's project instructions file and appending unsolicited content to it can surprise them on every subsequent agent turn.
- **Claude Desktop / Claude.ai**: save to a Project's context or a pinned note.
- **Cursor**: `.cursor/rules/business-context.md`.
- **Any other agent**: `seo-context.md` in the working directory.

Always confirm the write location with the user before saving. If any of these are unavailable or the user objects, fall back to re-asking the questions each session.

**Every use after that**: load that file first. If missing, ask where the user saved it or re-ask the questions.

Questions on first use:
- What does the business do, and who is it for?
- What's the brand's tone of voice? (Professional / Casual / Technical / Authoritative / Conversational)
- What language should content be written in?
- What topics should NEVER appear? (compliance, competitor mentions, etc.)
- Who are the 2-3 main competitors?

## Phase 1: Research

If a content brief wasn't provided, Google the topic and read the top 5 results. Note: what formats are ranking, what angles exist, what gaps you see. 3-5 bullet points, not a full brief.

Skip this phase entirely if a brief or prior conversation context already contains SERP analysis.

## Phase 2: Content Type Decision

Based on what's ranking, pick a content type: how-to, definition/explainer, comparison (X vs Y), listicle/roundup, product review, case study, pillar/ultimate guide, FAQ, landing page, service page, news/trend analysis.

State it plainly:
"The top results for [keyword] are all [format]. I'll write a [content type] with [key structural element]. Sound good, or did you have something else in mind?"

Wait for confirmation.

Load `references/content-types-overview.md` for the decision table covering all 23 content types. Then load the specific template from `references/content-types/<type>.md` (e.g., `references/content-types/how-to.md`) for H1/H2 structure, schema, featured snippet format, CTA placement, word count targets. The 19 content types bundled as full templates: how-to, definition, comparison, listicle, pillar-page, faq-page, landing-page, service-page, case-study, statistics-page, news-article, glossary-page, alternatives-page, buying-guide, product-page, category-page, integration-page, location-page, programmatic-page. For the 4 types covered only by the overview table (thought-leadership, product-reviews, pricing-pages, about-pages), those live under `eeat-audit/references/content-types/` because the E-E-A-T bar for them is the load-bearing factor.

## Phase 3: Knowledge Extraction

Ask 2-3 quick questions to extract unique knowledge the user has. Pick from:

- "What do most people