--- name: linkedin-writer description: LinkedIn-native long-form article and newsletter writing workflow for LinkedIn and Google-to-LinkedIn topic discovery, business-depth research, professional thought leadership, evidence-led drafting, final humanization, discussion design, SEO settings, auditing, and publish-ready packaging. Use when creating, outlining, researching, enriching, rewriting, humanizing, auditing, or packaging LinkedIn Articles, LinkedIn newsletters, LinkedIn long-form posts, LinkedIn thought leadership, LinkedIn B2B articles, LinkedIn 长文, LinkedIn 专栏, LinkedIn 话题调研, LinkedIn 商务内容, 去 AI 化编辑, or LinkedIn 发布包. ---

# LinkedIn Writer

## Goal

Turn a topic, product, argument, report, source bundle, or existing draft into a credible LinkedIn-native long-form article that helps a defined professional audience make a decision, understand a change, or improve how they work.

This skill reuses the parent `writer` workflow for fact checking, humanization, image packaging, and optional Cloudflare R2 delivery, but it does not treat LinkedIn as a generic SEO blog host or as Medium with a different publishing button.

LinkedIn-native writing prioritizes:

- a specific professional reader and work context; - a defensible point of view or useful decision framework; - current LinkedIn search and conversation signals; - expertise demonstrated through evidence, examples, and boundaries; - short, skimmable sections for busy readers; - a discussion-worthy close rather than a generic sales conclusion; - a complete native publishing pack, including LinkedIn SEO settings.

Default language follows the user's request. If the user gives no language, use the language of their source material or target audience.

## Format Routing

Use the requested format, not a blended default:

| Destination | Workflow | |---|---| | LinkedIn Article or LinkedIn newsletter edition | Use this skill in full | | LinkedIn short feed post only | Use the short-post rules and publishing pack in this skill; do not force a long article | | Google-first website article | Use `../SKILL.md` | | Medium article or third-party editorial essay | Use `../medium-writer/SKILL.md` | | Chinese WeChat Official Account article | Use `../wechat-writer/SKILL.md` |

If the user says only “LinkedIn article” or “LinkedIn long-form,” default to a native LinkedIn Article. If they already run a newsletter and provide its name or theme, package the piece as a newsletter edition. Do not claim a newsletter was created or published without direct evidence.

## Required References

Read each relevant file completely before acting:

- Topic discovery, LinkedIn search, trend expansion, and current seed topics: `references/linkedin-topic-research.md` - Google-to-LinkedIn discovery, business-depth enrichment, and final LinkedIn humanization: `references/linkedin-business-depth-and-humanization.md` - New article, rewrite, or reusable output format: `references/linkedin-article-template.md` - Article review, scoring, and revision gate: `references/linkedin-review-rubric.md` - Fact-heavy claims, comparisons, current products, or statistics: `../references/fact-check-and-style.md` - Final natural-language edit after factual and structural fixes: `../references/humanization.md` - Images, local paths, file packaging, and optional R2 delivery: `../references/output-packaging.md` - R2 upload tasks only: `../references/r2-image-upload.md` and `../references/r2-security.md`

Do not load unrelated references merely because they exist.

## Non-Negotiable Boundaries

1. Do not invent professional experience, product testing, customers, interviews, internal data, quotes, results, credentials, or events. 2. First-person events may appear only when the user supplied them for this task or they exist in an approved, attributable source package. 3. Do not turn LinkedIn search result counts, reactions, comments, or repeated phrases into search-volume claims. 4. Do not call a topic “trending,” “viral,” or “hot” without dated evidence. Use “recurring conversation,” “current topic seed,” or similarly bounded language when evidence is directional. 5. Do not mention or tag people and Pages merely to trigger notifications. Every suggested mention must have a content reason. 6. Do not convert a product announcement into disguised thought leadership. State affiliations, recommendations, and commercial relationships when they materially affect trust. 7. Do not use engagement bait such as “Agree?”, forced polls, empty controversy, or unrelated hashtags. Invite a concrete professional response. 8. Do not copy another LinkedIn creator's hook, framework, story, examples, distinctive phrases, or conclusion. Extract only topic signals and questions, then synthesize an original angle. 9. Do not claim publication, indexing, newsletter delivery, reach, or engagement from a completed local package. 10. Writing, generating images, uploading assets, and publishing externally are separate permission levels.

## LinkedIn Article Task Card

Before research or writing, create or infer a task card in 14 lines or fewer and save it as `linkedin-brief.md`:

- Publish as: personal profile / Company Page / unknown. - Format: standalone Article / newsletter edition / short feed post. - Professional audience: role, seniority, industry, and work situation. - Reader decision: what they should understand, compare, decide, or do. - Core thesis: one sentence the article must establish. - Expertise basis: supplied experience, verified sources, product knowledge, or editorial analysis. - Primary topic phrase: one natural phrase for LinkedIn and external search. - Related topic cluster: 4-8 entities, skills, problems, roles, or outcomes. - Conversation tension: trade-off, change, misconception, or unresolved question. - Business context: stakeholders, buying/approval path, economics, implementation, risk, and measurement dimensions that matter. - Evidence requirement: 3-6 claims that must be checked. - Target length: normally 900-1,800 words; adjust to the subject, not a platform myth. - CTA: discussion question, practical next step, subscription prompt, or disclosed product action.

Make conservative assumptions when details are missing. Ask only when audience, thesis, or authority to use personal experience is materially ambiguous.

## Working Modes

### Continuous mode (default)

Run `brief -> LinkedIn and Google-to-LinkedIn research -> business insight map -> evidence -> outline -> draft -> audit -> rewrite -> final humanization -> integrity recheck -> package` without pausing at every step. A request to “write an article” means deliver the reviewed local package.

### Topic-radar mode

When the user asks for hot topics, search ideas, or content planning, stop after the ranked topic map unless they also ask for an article. Do not draft ten shallow articles.

### Interactive mode

Pause at the topic shortlist or outline only when the user explicitly asks to choose first.

### Audit mode

If the user asks only for review, produce findings without overwriting the draft. If they ask to improve, preserve the original and apply fixes before re-auditing.

## End-to-End Workflow

### 1. Create an isolated article directory

Use:

```text writer/linkedin-writer/output/<article-slug>/ ```

This LinkedIn-specific output directory overrides the parent writer's default `writer/output/<article-slug>/` location. Prefer a short ASCII, hyphen-separated slug under 80 characters. Do not mix multiple campaigns in one directory.

Recommended working files:

```text linkedin-brief.md linkedin-topic-map.md linkedin-insight-map.md source-ledger.md outline.md draft.md article-linkedin.md linkedin-audit.md linkedin-publishing-pack.md image-plan.md ```

Create only the files the task needs. Keep `draft.md` separate from `article-linkedin.md` so an unreviewed draft cannot be mistaken for final copy.

### 2. Research LinkedIn search demand and conversation context

Read `references/linkedin-topic-research.md` completely.

Do not begin with a static list of broad trends. Build a query grid around:

```text core entity or skill × audience or role × work outcome × tension or decision × current change or timeframe ```

Use LinkedIn search suggestions and Posts results when available. Filter by recent date, content type, author industry/company, or source type when useful. Triangulate recurring questions with primary reports, official product or policy sources, credible industry research, customer questions, and the user's own content goals.

Then use Google to discover publicly indexed LinkedIn material with exact phrases, date operators, exclusions, and scoped queries such as:

```text site:linkedin.com/posts "<topic>" "<role or objection>" after:YYYY-MM-DD site:linkedin.com/pulse "<topic>" "<implementation, ROI, risk, or governance>" site:linkedin.com/company "<topic>" "<official case or report>" ```

Record Google results separately. Search snippets and LinkedIn creator claims are conversation signals, not automatically verified facts. Open the original page when possible, verify material claims elsewhere, and never copy a creator's hook, framework, structure, anecdote, or conclusion.

Record the exact query, date, filter, observed signal, and interpretation in `linkedin-topic-map.md`. Separate:

- **Observed:** directly visible search suggestion, repeated topic, question, format, or source. - **Inferred:** a possible reader need or angle derived from the observations. - **Verified demand:** use this label only when reliable demand data actually supports it.

Never imply that a topic is popular merely because it appears in one post or one search result.

### 3. Expand the topic before outlining

For the selected topic, create a useful professional topic cluster:

- Core concept: the named tool, skill, market shift, or decision. - Business outcome: time, quality, growth, cost, risk, hiring, retention, or customer value. - Role impact: what changes for practitioners, managers, executives, buyers, or candidates. - Implementation: workflow, prerequisites, governance, measurement, and failure modes. - Trade-off: what the popular framing misses or where the approach breaks. - Evidence: current data, official documentation, case material, or observable examples. - Adjacent conversation: 3-5 related topics that deepen the article without causing drift. - Discussion gap: a question qualified readers can answer from experience.

Reject adjacent topics that do not strengthen the thesis or reader decision. “More keywords” is not the same as more depth.

Read `references/linkedin-business-depth-and-humanization.md` and create `linkedin-insight-map.md`. Enrich the selected topic across the dimensions that materially affect the business decision:

- decision trigger and cost of waiting; - sponsors, users, approvers, blockers, buyers, and owners; - cost, budget, ROI, revenue, margin, or option value; - workflow, data, integration, adoption, and change management; - baselines, leading indicators, outcome metrics, and guardrails; - risk, strongest objection, failure mode, and reversibility; - one attributable or explicitly hypothetical scenario; - the next artifact, meeting, pilot, or decision the reader should initiate.

For a substantial business article, normally develop at least five relevant dimensions. Do not force irrelevant finance or governance sections into a career essay, but do not omit a material stakeholder, cost, or risk merely to keep the article simple.

### 4. Build the claim-source ledger

Create `source-ledger.md` for any current, factual, comparative, or decision-shaping article.

For each material claim, record:

- claim ID and exact claim; - claim type: fact / inference / editorial judgment / user-provided experience; - source title, publisher, date, and URL; - status: `verified`, `user_provided`, `needs_verification`, `softened`, `removed`, or `unsupported`; - where it will appear; - caveat or expiry risk. - rese