메뉴
Write a comprehensive article section with strong narrative prose, real examples, and inline citations. Reads word targets and prose ratio from article.json. Prioritizes flowing text over bullet lists.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | write-section |
| description | Write a comprehensive article section with strong narrative prose, real examples, and inline citations. Reads word targets and prose ratio from article.json. Prioritizes flowing text over bullet lists. |
| allowed-tools | Read, Write, Edit, Glob, WebFetch |
Write the section: $ARGUMENTS
Narrative over lists. Readers come for insight, not bullet points. Every list needs surrounding prose that explains the "why" and connects ideas. Lists are punctuation, not the main text.
Critical thinking over information dumps. Don't just present facts—analyze them. Ask: Why does this matter? What's the trade-off? When would you NOT do this? The best sections make readers think, not just note-take.
Headers that flow. Avoid double-barreled titles with colons like "Finding Prospects: Where to Look and Who to Target". Instead, use direct, punchy headers: "Where the Good Prospects Hide" or "Building Your First 100 Contacts".
Load Section Context
article.json for section scaffold, article_type, and content_typeword_target and word_minimum — these are the targets for THIS sectioneditorial_standards.prose_ratio_minimum — this is the prose ratio targetnarrative_role (hook, context, foundation, deep-dive, practical, inspiration, action, etc.)Load Research
research/[section-id].mdReview Previous Sections (if not first)
output/article.md for voice continuityAll targets come from article.json — do NOT use hardcoded values:
word_target fieldword_minimum field (fallback to quality_gates.min_words_per_section)editorial_standards.prose_ratio_minimumrequired_elements.examples_minimumrequired_elements.external_links_minimum (fallback to quality_gates.min_links_per_section)Adjust your writing approach based on the content_type in article.json:
Email/Newsletter: Get to the point fast. Short paragraphs. One clear CTA. Conversational and direct. No lengthy preambles.
News: Inverted pyramid — most important info first. Tight prose. Quote sources. Stay factual and timely.
Listicle: Each item gets a punchy header and supporting prose. Lists are expected but each item still needs narrative context.
Tutorial: Step-by-step clarity. Code samples with context. "Here's what we're building" before diving in. Check comprehension between steps.
Opinion: Strong thesis statements. Acknowledge counterarguments. Build an argument, not just a list of claims.
Reddit: Write like a real person posting to a subreddit. First person. No corporate voice, no buzzwords, no "leverage" or "game-changer" anywhere. Be direct, honest, and specific. Acknowledge limitations and downsides — reddit will call you out if you don't. Skip formal headers unless the post is long enough to need them. When you use headers, keep them dead casual. No emoji spam. The tone is "someone who actually knows what they're talking about sharing it plainly."
Long-form/Research/Case-study: Full narrative treatment as described in the writing process below.
Start mid-action or with a concrete scenario. Never start with "In this section, we'll cover..."
Scale the opening length to the section size:
Strong openings:
The first paragraph should make the reader lean in.
Write in prose paragraphs, not bullet cascades.
When you have 5 related points, don't write 5 bullets. Write 2-3 paragraphs that weave those points into a coherent argument. Use a list only when you're presenting truly parallel items (like a tool comparison or a step-by-step checklist).
The prose-to-list ratio: Check editorial_standards.prose_ratio_minimum from article.json. If the target is 70%, aim for 70% flowing prose and 30% structured elements. If the target is 40% (listicle), lists are more acceptable but still need surrounding context.
Pattern for deep explanations:
Example of prose over bullets:
DON'T do this:
Benefits of follow-ups:
- Increases response rates by 21%
- 70% of responses come from emails 2-4
- Shows persistence
- Gives multiple chances to catch them at the right time
DO this:
Follow-ups aren't optional—they're where deals actually close. The data
is striking: a single follow-up increases response rates by 21%, and
three follow-ups can push that number to 50%. Most responses don't
come from your first email at all. According to research from Lemlist,
70% of positive responses arrive on the second, third, or fourth touch.
The psychology makes sense. Decision-makers are buried in email. Your
first message might arrive during a crisis, a meeting, or simply a
bad morning. Following up isn't pestering—it's giving them another
chance to engage when the timing is better.
Weave examples into the narrative. Don't isolate them in boxes.
Good example integration:
Take Marco Massaro's campaign. He targeted smaller companies where he'd
work directly with decision makers, personalized each email with specific
observations about their websites, and implemented systematic follow-ups.
His results? A 9% initial response rate that jumped to 13% after follow-ups—
and a $15,000 consulting project closed within three months.
The key insight from Marco's approach isn't volume. It's precision. He
didn't blast 5,000 generic emails. He sent 300 thoughtful ones to the
right people.
When including technical evidence (code, data tables, etc.):
Introduce with context explaining what the reader is about to see.
[evidence block]
Follow with analysis explaining what it means and why it matters.
Avoid:
Use instead:
Headers should sound like something a smart friend would say, not like academic paper sections.
End with momentum. Bridge to what's next without the phrase "In the next section..."
Scale closing length to section size:
Strong closings:
Before finishing, check against article.json targets:
word_minimumeditorial_standards.prose_ratio_minimumSection headers (## level): Short, punchy, curiosity-inducing. 3-7 words ideal.
Subsection headers (### level): Can be slightly longer but should still flow conversationally.
Use lists for:
Don't use lists for:
When you do use lists:
output/article.mdWrite in final publication format with natural headers.
Mark section as written with word count and link count.
Log completion with quality metrics.
Section FAILS if:
word_minimum (from article.json)editorial_standards.prose_ratio_minimumReport:
/write-section [next] or /embed-mediaRemove signs of AI-generated writing from text. Use when editing or reviewing text to make it sound more natural and human-written. Based on Wikipedia's comprehensive "Signs of AI writing" guide. Detects and fixes patterns including: inflated symbolism, promotional language, superficial -ing analyses, vague attributions, em dash overuse, rule of three, AI vocabulary words, negative parallelisms, and excessive conjunctive phrases.
Discover and scrape jobs from 2,800+ companies using Ashby as their ATS. Search and filter by location, remote status, department, and keywords. Use when looking for jobs, researching companies hiring, or building job market intelligence.
Scrape all Ashby job boards and search for matching roles. Runs the full discover-scrape-search pipeline. Use when user says "scrape ashby" or wants to run the full job scraping pipeline.
Professional fiction editor with 6 passes. Checks AI pattern contamination, voice fidelity, POV consistency, continuity, prose tightening, and scene assessment. Assembles scenes into chapter.md.
Add YouTube video embeds and mermaid diagrams to the article. Replaces placeholders with actual embed code.
Evaluate article quality against targets defined in article.json. All thresholds are read from the spec, not hardcoded. Returns PASS or FAIL with specific issues.