Social: LinkedIn & the realistic 2026 picture

The honest channel map

The target: a week of posts queued as dated files in approved/social/ - five drafts from one of your real ideas, the survivors picked and edited by you, final-formatted for LinkedIn and X with spaced publish dates. Getting there starts with honesty about the platforms. Most social-automation advice online is either outdated or pretends the platform rules don't exist. Here is the map as it actually stands, so you build on ground that holds.

• LinkedIn: automated posting is prohibited outside approved partner programs, and there is no sanctioned self-serve path. Anything that posts for you through a headless browser or a scraped session is account-ban territory.
• X: programmatic posting exists but is pay-per-use, terms-restricted, and repriced often enough that building on it is rented ground for an SMB.
• The pattern across every network: the platforms want humans pressing post, and they change the rules for everything else quarterly.

The conclusion writes itself: do all the work in Claude, end at finished, dated post files, and treat the actual posting as a tiny daily ritual - or hand the dated files to whatever scheduler you already trust, a one-line handoff that changes nothing upstream.

The queue is a folder

The queue lives at content/approved/social/ - one file per post, named by date and channel. The file is the finished artifact: final text, channel, publish date, all in one place. Whoever posts (you, a teammate, or a scheduler you already use) just works through the folder in date order.

content/approved/social/
├── 2026-06-23_linkedin_pricing-story.md
├── 2026-06-24_x_pricing-numbers.md
├── 2026-06-26_linkedin_contrarian-take.md
└── ...

# inside each file:
---
channel: linkedin
publish_date: 2026-06-23
status: approved
---
We spent $4,200 to learn this.

You get it for free...

A small skill - queue-view - reads the folder and renders this week's schedule as one page: what posts when, on which channel, and which days are still thin. Claude Code writes it in minutes; the point is that scheduling becomes a property of files, not a tab someone keeps open.

Topics come from the live feed, the substance from your files

The fastest way to write posts nobody reads is to mine your own repo for topics. We tried it: 20 topics pulled from our own builds and calls, all true, all ours, and all rejected, because nobody in the feed that week was arguing about any of them. The fix is to split the two jobs. The feed tells you what people are talking about this week. Your corpus tells you what you can say about it that nobody else can.

1. Pull what's live: the top posts in your niche from the last 7 days, via a daily news digest (next module) or a scrape you've priced and approved. Save them as files in data/feed/.
2. Have Claude group them into waves: the 3-5 conversations the feed is actually having, with the posts driving each.
3. For each wave, ask what you own that nobody else in that conversation has: a number, a build, a call, a mistake.
4. Post the inventory version, with both sides. When everyone posts that they cancelled a tool, post what you replaced and what you kept. The kept column is the credibility. One number is a brag; yours next to the average next to the best is a resource.
5. Write first-person work (what I built, what I tested, what I decided), not market commentary. 'X is dying' posts are generic by construction.

Then fix the week so the engine knows what to fill. A cadence is a set of slots with a format each, not a posting frequency. One that works for a founder account: Monday a lead magnet, Tuesday and Thursday a data or list post built from something you measured, Wednesday and Friday personal (a mistake, a decision, a story). Weekends empty. Write it into calendar/cadence.md and have the planning skill fill the slots from the topic backlog every Sunday.

One idea, a week of posts

Now the production recipe this all serves: one good idea becomes five scheduled posts, each a genuinely different post rather than five rewordings.

1. Start from one idea - a moment from your repurposing folder, a take from a call, a number from your own data.
2. Run the write-linkedin skill: it generates five angles - story, contrarian, how-to, numbers, list - and drafts each one against voice.md and the LinkedIn format rules.
3. Drafts land in content/drafts/. You read all five, edit the keepers, and kill the weak ones. Three survivors out of five is a good week.
4. Move survivors to content/approved/social/ with spaced publish dates in the filename - that move is the approval.
5. Post from the queue each morning, or in one weekly batch. Anything still in the folder can be pulled or edited right up to the moment it goes out.

---
name: write-linkedin
description: Turn one idea into 5 LinkedIn post drafts using the five
  angles (story, contrarian, how-to, numbers, list). Loads
  brand/voice.md and the LinkedIn format rules. Writes to
  content/drafts/ only - never schedules, never posts.
argument-hint: <idea or moment-file>
model: sonnet
---

The model: sonnet line is the routing decision: five variants are volume work, and you'll kill two of them anyway. The survivors still go through the editor subagent on Opus and the lint script before they reach approved/social/.

The approval gate, applied to social

The same folder doctrine carries over: skills write to drafts/, the queue holds only what a human moved into approved/social/, and a human is the only thing that moves files between them. No approved file, no post - the queue physically can't contain what you haven't approved.

And the boundary worth restating: nothing in this workflow bots LinkedIn itself. No headless browsers posting on your behalf, no scraped-session tools - that's account-ban territory and the accounts at risk are your founder's and your company's. Write in the repo, post like a human, and sleep fine.

Do this now

Sources and further reading

• Claude Code skills docs
• How Anthropic's marketing team uses Claude