Skip to content
Back to search

linkedin-company-page-skills

MarketplaceWorks with Claude CodeGitHubIncomplete

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked yesterday
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/chandewardnyanesh--linkedin-company-page-skills/badge.svg)](/listings/chandewardnyanesh--linkedin-company-page-skills)

HTML

<a href="/listings/chandewardnyanesh--linkedin-company-page-skills"><img src="/api/listings/chandewardnyanesh--linkedin-company-page-skills/badge.svg" alt="findsafeskills safety rating" /></a>

Install

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

Claude Code (plugin)

git clone https://github.com/chandewardnyanesh/linkedin-company-page-skills.git
cd linkedin-company-page-skills
python -m scripts.selftest      # 19 checks, offline, no dependencies

Then add it as a plugin marketplace in Claude Code, pointing at the cloned directory — the repo ships `.claude-plugin/marketplace.json` and `.claude-plugin/plugin.json`.

Claude Desktop or claude.ai

Upload any individual `skills/<name>/SKILL.md` as a skill, or the whole directory as a plugin where your plan supports it. Each skill works standalone; they hand off to each other by name when both are present.

Requirements

**None.** `lib/` is Python standard library only and the default path makes no network call. Python 3.10+ for the helper scripts.

**Optional, used when present:** Google Search Console, Ahrefs or Semrush connectors sharpen `trend-radar` and `page-performance-report` with measured search demand instead of inference. Without them, those skills label their evidence as observed rather than measured and say so.

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

.claude-plugin/marketplace.json · Manifest · 27 lines
  1. {
  2. "name": "linkedin-company-page-skills",
  3. "metadata": {
  4. "description": "Claude skills for LinkedIn company Pages: brand profile research, Page optimisation, trend and competitor intelligence, multi-surface content calendars, Page posts and document carousels, authority articles and newsletters, de-AI and de-corporate humanizer, repurposing, proof extraction, employee amplification, comment desk, monthly reporting."
  5. },
  6. "plugins": [
  7. {
  8. "name": "linkedin-company-page-skills",
  9. "source": "./",
  10. "description": "13 single-purpose skills for running a LinkedIn company Page, built on the 2026 reality that a Page reaches 1-2% of its followers and has to borrow distribution from people. Includes a claim-gated writing path (no figure publishes without a cleared proof row), a humanizer that strips press-release voice as well as AI tells, an employee amplification layer with per-person angles rather than one shared caption, and honest handling of the surfaces LinkedIn exposes no API for.",
  11. "version": "1.0.1",
  12. "license": "MIT",
  13. "keywords": [
  14. "linkedin",
  15. "company-page",
  16. "b2b-marketing",
  17. "brand-voice",
  18. "content-strategy",
  19. "employee-advocacy",
  20. "thought-leadership",
  21. "newsletter",
  22. "claude-skills"
  23. ],
  24. "category": "productivity"
  25. }
  26. ]
  27. }
README.md · README · 375 lines
  1. # LinkedIn Company Page Skills for Claude
  2. **13 Claude skills for running a LinkedIn company page in 2026** — website
  3. research, content strategy, page post writing, authority articles and
  4. newsletters, a de-AI and de-corporate humanizer, employee advocacy, and
  5. monthly analytics reporting.
  6. Built for the one fact most LinkedIn advice ignores: **a company page reaches
  7. 1–2% of its own followers.** Everything here follows from that.
  8. [![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](LICENSE)
  9. [![Skills: 13](https://img.shields.io/badge/skills-13-brightgreen.svg)](#the-13-skills)
  10. [![Dependencies: none](https://img.shields.io/badge/dependencies-none-brightgreen.svg)](#installation)
  11. [![Claude Code](https://img.shields.io/badge/Claude%20Code-compatible-8A63D2.svg)](https://claude.com/claude-code)
  12. Works with **Claude Code**, **Claude Desktop** and the **claude.ai** skills
  13. interface. No API keys, no signup, no Python packages.
  14. ---
  15. ## Contents
  16. - [Why company pages need their own skills](#why-company-pages-need-their-own-skills)
  17. - [Installation](#installation)
  18. - [The 13 skills](#the-13-skills)
  19. - [How a request flows](#how-a-request-flows)
  20. - [Three design decisions](#three-design-decisions)
  21. - [What LinkedIn's API can and cannot publish](#what-linkedins-api-can-and-cannot-publish)
  22. - [Repository layout](#repository-layout)
  23. - [FAQ](#faq)
  24. - [What this project does not claim](#what-this-project-does-not-claim)
  25. - [Credits](#credits)
  26. ---
  27. ## Why company pages need their own skills
  28. A LinkedIn company page is **not a reach channel**. It is an authority and
  29. conversion asset whose distribution has to be borrowed from people. The 2026
  30. data, from several independent directions:
  31. | Measure | Company page | Personal profile | Source |
  32. |---|---|---|---|
  33. | Median engagement rate | 1–2% | ~4.7% | Sprout Social Q1 2026 Index, 52M+ posts |
  34. | Share of feed distribution | ~5% | ~65% | SocialPilot, Sept 2026 |
  35. | Organic reach vs follower count | ~1.2% | 8–15% | SMM News 2026, 524 B2B brand pages |
  36. | Reach change, 2024 → early 2026 | **down 60–66%** | down ~50% | multiple 2026 roundups |
  37. | Median reactions per B2B post | 23 | 39 (founder/exec) | Imagine AI, Oct 2026 |
  38. | Reach per post vs employee posts | baseline | **+561%** | SocialPilot, Sept 2026 |
  39. So a page with 20,000 followers can reach fewer people than a founder with
  40. 2,000. Any content plan whose growth depends on page post volume does not work.
  41. **This is also why a personal-brand skill bundle cannot simply be pointed at a
  42. company page.** The personal formula libraries are built on devices a brand
  43. cannot use honestly — a confession hook needs someone to confess, and a company
  44. has no self. The attempt produces the worst genre on the platform: a corporate
  45. entity performing feelings.
  46. What a page has instead is **access a person does not have**: the whole book of
  47. business, the aggregate numbers, the thing that broke in production, the pattern
  48. across forty clients, the real cost of the work. A person can report their own
  49. experience. A page can report a population. Every format in this repository is
  50. a way of spending that advantage.
  51. The four jobs a company page actually does well, and which every skill here
  52. serves:
  53. 1. **Pass the credibility check** — the buyer who met you elsewhere and opened
  54. the page before replying.
  55. 2. **Hold the assets that get borrowed** — carousels, articles, research that
  56. employees reshare.
  57. 3. **Own distribution the feed cannot throttle** — a LinkedIn newsletter
  58. notifies subscribers by email.
  59. 4. **Convert** — custom button, lead-gen forms, products and services tabs.
  60. Full sourcing and the honest reading of every figure:
  61. [`references/page-algorithm-heuristics.md`](references/page-algorithm-heuristics.md).
  62. ## Installation
  63. ### Claude Code (plugin)
  64. ```bash
  65. git clone https://github.com/chandewardnyanesh/linkedin-company-page-skills.git
  66. cd linkedin-company-page-skills
  67. python -m scripts.selftest # 19 checks, offline, no dependencies
  68. ```
  69. Then add it as a plugin marketplace in Claude Code, pointing at the cloned
  70. directory — the repo ships `.claude-plugin/marketplace.json` and
  71. `.claude-plugin/plugin.json`.
  72. ### Claude Desktop or claude.ai
  73. Upload any individual `skills/<name>/SKILL.md` as a skill, or the whole
  74. directory as a plugin where your plan supports it. Each skill works
  75. standalone; they hand off to each other by name when both are present.
  76. ### Requirements
  77. **None.** `lib/` is Python standard library only and the default path makes no
  78. network call. Python 3.10+ for the helper scripts.
  79. **Optional, used when present:** Google Search Console, Ahrefs or Semrush
  80. connectors sharpen `trend-radar` and `page-performance-report` with measured
  81. search demand instead of inference. Without them, those skills label their
  82. evidence as observed rather than measured and say so.
  83. ## The 13 skills
  84. ### Setting up
  85. | Skill | What it does |
  86. |---|---|
  87. | **`page-profile-builder`** | Crawls the company website and LinkedIn page, writes the canonical brand profile, proof library, ICP and competitor set every other skill reads. **Run this first.** |
  88. | **`page-optimizer`** | Audits and rewrites the page itself — tagline (120 chars), About (2,000 chars), specialties, custom button, Products and Services tabs, Featured, Life tab |
  89. ### Deciding what to say
  90. | Skill | What it does |
  91. |---|---|
  92. | **`trend-radar`** | Ranked content opportunities with graded evidence. Uses Search Console / Ahrefs / Semrush when available |
  93. | **`competitor-page-analyzer`** | Competitor cadence, format mix, newsletters, employees posting — and the white space you have standing to own |
  94. | **`page-content-calendar`** | 30-day multi-surface plan built on real capacity, not a posting target |
  95. ### Writing
  96. | Skill | What it does |
  97. |---|---|
  98. | **`page-post-writer`** | LinkedIn page posts across 16 brand formats, picked by the job the post has to do |
  99. | **`authority-article-writer`** | LinkedIn Articles and newsletter issues across 8 authority architectures |
  100. | **`page-humanizer`** | Strips AI tells **and** press-release voice. `--mode audit` runs a six-gate pre-publish review |
  101. ### Reusing what exists
  102. | Skill | What it does |
  103. |---|---|
  104. | **`page-repurposer`** | One source into several native assets — blog to carousel, webinar to posts, founder's post to page version |
  105. | **`page-asset-extractor`** | `--mode pattern` deconstructs any high-performing post · `--mode proof` mines documents for claims |
  106. ### Distribution and inbound
  107. | Skill | What it does |
  108. |---|---|
  109. | **`employee-amplification`** | Per-person reshare kits with distinct angles, the once-a-week notify budget, Thought Leader Ad selection. **This is the distribution layer** |
  110. | **`page-engagement-desk`** | Comment replies, complaint triage and escalation, commenting as the page |
  111. ### Measuring
  112. | Skill | What it does |
  113. |---|---|
  114. | **`page-performance-report`** | Monthly report that says what to change, measured against your own baseline rather than borrowed benchmarks |
  115. ## How a request flows
  116. 1. **Load the profile.** Nothing is written from a blank slate.
  117. 2. **Draft**, using the formats, architectures and voice rules in
  118. [`references/`](references/).
  119. 3. **Show an approval card** — every claim with its source, the surface
  120. character fit, the amplification plan, and what happens on approval.
  121. 4. **Publish only after a human approves that exact text.**
  122. ## Three design decisions
  123. ### 1. No figure publishes without a cleared proof row
  124. A number on a company page is a representation the business made, in public,
  125. with a timestamp, to an audience that includes its customers and competitors.
  126. It gets screenshotted before it gets corrected.
  127. So every writing skill stops at a claim with no entry in
  128. [`references/proof-library.md`](references/proof-library.md). A format's
  129. `[specific number]` slot is **not** a licence to produce a plausible one —
  130. plausible is fabricated. When the library has nothing that fits, the skill
  131. asks, drops the claim, or changes format. It does not soften an unevidenced
  132. number into "up to".
  133. This is the behaviour most LLM marketing tooling gets wrong, and it is the
  134. single most expensive thing to get wrong on a brand account.
  135. ### 2. The humanizer fixes two voices, not one
  136. Brand copy sounds like AI **and** like a press release, and the two reinforce
  137. each other because they share most of their tells. Two tests on every opening
  138. line:
  139. - **Swap in a competitor's name.** Survives? It says nothing. → press-release voice
  140. - **Swap in a different industry.** Survives? Cut it. → AI voice
  141. Both resolve the same way, and it is not word substitution — "utilise" for
  142. "leverage" is the same empty slot with a different word in it. The fix is a
  143. named person, a specific number, or a dated event. Worked rewrites in
  144. [`references/brand-voice-rules.md`](references/brand-voice-rules.md).
  145. ### 3. Manual publishing is the honest answer, not a missing feature
  146. See the table below. For articles and newsletters there is no API to use, so
  147. "here is the draft, paste it into the editor" is complete and correct, and no
  148. skill here implies a key would change it.
  149. ## What LinkedIn's API can and cannot publish
  150. | Surface | Publishable by API? |
  151. |---|---|
  152. | Page post, document post, comment | Yes — `w_organization_social` scope + admin role |
  153. | **LinkedIn Article (Articles for Pages)** | **No. No publishing API exists** |
  154. | **LinkedIn Newsletter** | **No. An extension of articles, so no API either** |
  155. | Poll, Event | No |
  156. | Tagline, About, specialties, tabs, Featured | No |
  157. Any tool claiming to automate a LinkedIn article is driving a browser or is
  158. wrong.
  159. This bundle ships **draft-first on every surface**. To automate the
  160. API-reachable ones, export a poster command in a trusted shell:
  161. ```bash
  162. export LINKEDIN_PAGE_CUSTOM_POSTER='/path/to/your-poster'
  163. ```
  164. Approved drafts are piped to it as JSON on stdin. The variable is
  165. **deliberately ignored inside `.env` files** — a file on disk choosing which
  166. executable runs is a code-execution hole dressed as configuration.
  167. ## Repository layout
  168. ```
  169. linkedin-company-page-skills/
  170. ├── SKILL.md # router: which skill for which request
  171. ├── skills/ # 13 skills, each a single-purpose SKILL.md
  172. ├── references/ # 14 shared reference files (see below)
  173. ├── lib/ # standard library only, no network calls
  174. │ ├── surfaces.py # every character limit + a draft checker
  175. │ ├── url_parser.py # page URLs → slug, numeric id, organization URN
  176. │ ├── approval.py # the approval card
  177. │ └── backend_selector.py # manual publish + shell-only extension point
  178. ├── scripts/ # fit, selftest, 3 validators
  179. ├── tests/ # handoff-contract integrity
  180. ├── evals/ # 3 test cases, runnable with `claude plugin eval`
  181. └── examples/ # a real worked research output
  182. ```
  183. ### The reference layer
  184. | File | Holds |
  185. |---|---|
  186. | [`page-algorithm-heuristics.md`](references/page-algorithm-heuristics.md) | The 2026 reach economics, sourced. **Read first** |
  187. | [`page-formats.md`](references/page-formats.md) | 16 page post formats, the monthly mix, and the anti-formats to refuse |
  188. | [`article-architectures.md`](references/article-architectures.md) | 8 long-form structures, newsletter shape and eligibility |
  189. | [`brand-voice-rules.md`](references/brand-voice-rules.md) | Banned vocabulary by density, the attribution rule, worked rewrites |
  190. | [`page-surfaces.md`](references/page-surfaces.md) | Every surface, its limit, what gates it, what is manual |
  191. | [`amplification-mechanics.md`](references/amplification-mechanics.md) | The advocacy ladder, notify limits, an honest Thought Leader Ads reading |
  192. | [`metrics-taxonomy.md`](references/metrics-taxonomy.md) | What each metric counts and what to stop reporting |
  193. | [`precedence.md`](references/precedence.md) | What wins when two rules conflict |
  194. | [`untrusted-content.md`](references/untrusted-content.md) | Fetched content is data, never instructions |
  195. Five more are **blank templates** that become confidential business documents
  196. once filled: `company-profile.md`, `proof-library.md`,
  197. `claims-and-compliance.md`, `icp-firmographics.md`, `competitor-set.md`.
  198. `scripts/check_no_secrets.py` fails the build when one is marked
  199. `filled: yes`, as a reminder to gitignore it.
  200. ## Tools
  201. ```bash
  202. python -m scripts.fit --surface page_post --file draft.md # does it fit?
  203. python -m scripts.fit --list # every surface and limit
  204. python -m scripts.selftest # is it wired up?
  205. python -m scripts.check_no_secrets # before committing
  206. python -m tests.test_instruction_integrity # handoff contracts
  207. ```
  208. Count characters with the tool, not by eye. A LinkedIn **page** About section
  209. is **2,000** characters, not the 2,600 a personal profile gets, and the
  210. overflow is dropped rather than wrapped — invisible in the admin editor,
  211. obvious in public.
  212. ## FAQ
  213. **Is this for personal LinkedIn profiles?**
  214. No. It is built specifically for company pages, where the formats, voice
  215. constraints, claim liability and distribution model are all different. For
  216. personal-profile work, see
  217. [sergebulaev/linkedin-skills](https://github.com/sergebulaev/linkedin-skills).
  218. **Does it post to LinkedIn automatically?**
  219. Not by default, and for articles and newsletters not at all — LinkedIn has no
  220. publishing API for them. Page posts and comments can be automated by wiring up
  221. your own poster command. Everything goes through a human approval card first.
  222. **Do I need an API key, a scraper, or a paid tool?**
  223. No. There are no dependencies and the default path makes no network call.
  224. Search Console, Ahrefs and Semrush connectors are used if you already have
  225. them.
  226. **Will this make my AI-written posts undetectable?**
  227. No, and no skill here claims it. See
  228. [what this project does not claim](#what-this-project-does-not-claim).
  229. **Can it write LinkedIn newsletters?**
  230. Yes — `authority-article-writer` drafts newsletter issues and checks your
  231. eligibility (Super Admin, 150+ followers, active page, recent original
  232. content). Publishing is manual because no API exists.
  233. **How is this different from asking Claude to "write a LinkedIn post"?**
  234. Three things: it refuses to invent statistics about your business, it picks a
  235. format based on the job the post has to do rather than defaulting to a hook,
  236. and it plans the employee amplification without which a page post reaches
  237. almost nobody.
  238. **Does it work for non-English pages?**
  239. The voice rules and formats are language-agnostic in structure, and
  240. `company-profile.md` records the publishing language. The banned-vocabulary
  241. lists are English; contributions for other languages are welcome.
  242. **Is it safe to point at my company website?**
  243. Yes. Everything fetched is treated as data, never as instructions, and a
  244. figure found on your own website is **not** treated as evidence until a human
  245. clears it — which is how stale claims get republished with fresh timestamps.
  246. See [`references/untrusted-content.md`](references/untrusted-content.md).
  247. ## What this project does not claim
  248. - **It does not beat AI detectors.** No edit reliably does, detector scores
  249. move between model generations, and claiming otherwise would be dishonest.
  250. What these rules do is make copy specific, attributable and checkable, which
  251. is what a reader responds to.
  252. - **It does not promise reach.** Every benchmark in the references measures
  253. someone else's corpus, most of it vendor data with a proprietary score
  254. behind it. Your own trailing baseline beats all of them.
  255. - **It does not attribute revenue to posts.** A company page cannot show which
  256. post produced a deal, and `page-performance-report` says so in every report
  257. rather than modelling what the data will not support.
  258. - **It does not automate LinkedIn articles or newsletters.** Nothing can.
  259. ## Contributing
  260. Issues and pull requests welcome — particularly reference-file corrections
  261. when LinkedIn changes a limit or a gate, and banned-vocabulary lists for
  262. languages other than English. See [CONTRIBUTING.md](CONTRIBUTING.md).
  263. Before opening a PR:
  264. ```bash
  265. python -m scripts.selftest
  266. python -m tests.test_instruction_integrity
  267. python -m scripts.check_markdown_references
  268. ```
  269. ## Credits
  270. Structure and conventions follow the pattern established by
  271. [sergebulaev/linkedin-skills](https://github.com/sergebulaev/linkedin-skills)
  272. (MIT) for personal-profile LinkedIn work: single-purpose skills, a shared
  273. reference layer with an explicit precedence order, an untrusted-content rule,
  274. and draft → approval → publish throughout.
  275. The formats, voice rules, claim gating, distribution model and surface
  276. handling in this repository are built for company pages, which behave
  277. differently.
  278. Built by [Dnyanesh Chandewar](https://github.com/chandewardnyanesh).
  279. ## Licence
  280. [MIT](LICENSE).
  281. ---
  282. <sub>Keywords: LinkedIn company page, LinkedIn business page, Claude skills,
  283. Claude Code plugin, LinkedIn content strategy, B2B LinkedIn marketing,
  284. LinkedIn employee advocacy, LinkedIn newsletter, LinkedIn document carousel,
  285. thought leadership, brand voice, AI content humanizer, LinkedIn analytics,
  286. company page SEO.</sub>

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.