article-writing

name	article-writing
description	Write articles, guides, blog posts, tutorials, newsletter issues, and other long-form content in a distinctive voice derived from supplied examples or brand guidance. Use when the user wants polished written content longer than a paragraph and voice consistency, structure, and credibility matter. Do NOT use for short social posts or threads (content-engine), for deriving the voice profile itself (brand-voice), or for search-brief work (seo).
origin	ECC

Article Writing

Write long-form content that sounds like a real person or brand, not generic AI output.

When to use

• drafting blog posts, essays, launch posts, guides, tutorials, or newsletter issues
• turning notes, transcripts, or research into polished articles
• matching an existing founder, operator, or brand voice from examples
• tightening structure, pacing, and evidence in already-written long-form copy

When NOT to use:

• the deliverable is a tweet, thread, or LinkedIn post — that is content-engine
• the job is to derive a reusable voice profile rather than write — that is brand-voice
• the required facts do not exist yet. Long-form without sourced material produces invented biography and invented metrics, which is the worst failure this skill has (see Failure modes)

Inputs

Required before drafting — absent input = stop and ask, never invent:

1. Audience and purpose — who reads it, what they should be able to do afterwards.
2. Source material for every factual claim — notes, transcripts, docs, research, or a live probe. For SkyyRose content: product facts come only from wordpress-theme/skyyrose-flagship/data/skyyrose-catalog.csv, collection language only from docs/brand/collection-stories.md, founder voice only from docs/brand/corey-questions.md. Never a biographical or metric claim from memory.
3. Voice reference — a VOICE PROFILE from brand-voice, or 3+ real samples (published articles, newsletters, X/LinkedIn posts, docs, memos, a style guide). If none exist, default to a direct operator voice: concrete, practical, low on hype — and say that is what you used.

Procedure

1. Confirm audience, purpose, sources, and voice reference.
2. Extract from the voice samples: sentence length and rhythm · formal vs conversational vs sharp · favored devices (parentheses, lists, fragments, questions) · tolerance for humor, opinion, contrarian framing · formatting habits (headers, bullets, code blocks, pull quotes).
3. Build a skeletal outline with one purpose per section.
4. Draft each section starting with the concrete thing — example, output, anecdote, number, described screenshot, or code block. Explain after the example, never before.
5. Expand only where the next sentence earns its place. Prefer short direct sentences.
6. Attach a source marker to every number and every factual claim as you write it ((Source: …)), so check 3 below can be run mechanically rather than by re-reading.
7. Apply the structure pattern for the format:
   • Technical guide — open with what the reader gets; a code or terminal example in every major section; end with concrete takeaways, not a soft summary.
   • Essay / opinion — start with tension, contradiction, or a sharp observation; one argument thread per section; examples that earn the opinion.
   • Newsletter — strong first screen; insight mixed with updates, not diary filler; clear section labels and easy skim structure.
8. Run the Verification checks on the draft file before delivering.

Verification

Write the draft to a real file so the checks have a target. grep exiting 2 means the file path is wrong — a dead gate, not a pass (bug-230). Exit 1 is clean; exit 2 is broken.

1. Banned-pattern scan — the tropes that make copy read as AI output:

grep -nEi 'rapidly evolving|game.chang|cutting.edge|revolutionary|moreover,|furthermore,|excited to share|in today.s .* landscape' article.md

PASS: exit 1, zero matches. Any hit gets deleted and rewritten, not softened. [repro]

2. Heading structure — one purpose per section starts with one H1:

grep -c '^# ' article.md

PASS: prints 1. Two H1s means two articles; zero means no stated subject. [repro]

3. Every number carries a source — the check that stops invented metrics:

grep -nE '[0-9]+(\.[0-9]+)?%|\$[0-9]' article.md | grep -v 'Source:'

PASS: exit 1, no lines returned. A number with no (Source: …) is either sourced now or deleted. operator-supplied; unverified is an acceptable source label; silence is not. [repro]

Prove check 3 can fail before trusting it (rule 3): drop one unsourced figure into the draft, confirm the grep returns it, restore. Observed doing exactly this in the worked example.

A SKIP is not a PASS: if voice fidelity cannot be checked mechanically (it cannot — no command grades tone), say so explicitly and name who closes it. The strongest available check is mechanical: banned-pattern scan (check 1) plus a side-by-side read of one drafted paragraph against a named source sample. The human requester closes the tone verdict, not the drafter.

Worked example

Real run, 2026-07-28. Draft written to the session scratchpad as article.md, subject: the one-run drop model, with the brand line from docs/brand/collection-stories.md [repo].

$ grep -c '^# ' article.md
1
$ grep -nEi 'rapidly evolving|game.chang|cutting.edge|revolutionary|moreover,|furthermore,' article.md; echo "exit=$?"
exit=1
$ grep -nE '[0-9]+(\.[0-9]+)?%|\$[0-9]' article.md | grep -v 'Source:'; echo "exit=$?"
exit=1

The unsourced-number gate was proven red first, on a sibling file in the same run:

$ grep -nE '[0-9]+(\.[0-9]+)?%|\$[0-9]' research-brief.md | grep -v 'Source:'
3:- Average luxury-streetwear AOV is $240

After appending (Source: operator-supplied; unverified) to that line, the same command exits 1. Both states observed this session [repro]. Passing the sourcing gate is what the article's 30% unit-cost premium (Source: operator-supplied) line exists to satisfy.

Dead-gate demonstration from the same run: grep -nEi 'anything' no-such-file.md printed no such file or directory and exited 2 — zero matches, but not a pass [repro].

Failure modes

Failure	What it looks like	Rule
Invented biography or metrics	"grew 4x in 18 months" with no source	Never invent biographical facts, company metrics, or customer evidence. Check 3 catches numbers; claims still need a human read
Hallucinated product/brand fact	wrong SKU, price, or collection story	bug-096 shape. Catalog CSV and docs/brand/collection-stories.md are the SOT; plausible ≠ sourced
Collection canon mixed	"bloodline" language outside Love Hurts	Each collection has its own canon; never substitute one collection's voice for another
Exit 2 read as clean	grep on a wrong path reported as "no banned patterns"	bug-230 fail-open. Exit 1 = clean, exit 2 = broken gate
Voice check silently skipped	delivered as "matches the voice" with no comparison	A SKIP is not a PASS. Name the sample compared against, or state the tone verdict is open and owned by the requester
Explanation before evidence	section opens with abstraction, example arrives late	Lead with the concrete thing; explain after
Filler transitions and hype	"Moreover", "game-changer", "cutting-edge"	Check 1 catches these mechanically — rewrite, do not soften