LinkedIn Post

Turns one idea into a post that sounds like the person who will publish it. The user posts it; this skill only writes.

Phase 1: Context

1. Read the Voice section of 01-Perfiles/Perfil Maestro.md in the career workspace (/master-profile). No voice yet: ask for three of the user's own past posts, infer it from them and offer to save it there. Never invent a voice.
2. Read 01-Perfiles/Personal Brand.md if /personal-brand wrote a plan: the slot gives the type, the angle and the hook number.
3. A thin idea ("post about AI") gets one batched question, not padding: what happened, to whom, and what did it cost or return. A post needs one specific true thing.
4. A topic to explain rather than something that happened ("explain the KV cache", "why physics goes in FixedUpdate") goes through Explain mode below before Phase 2.

Explain mode

For a post that teaches one concept from games, software or LLMs. The specific true thing is the example, not an anecdote.

1. One concept, one reader. Narrow the topic until it fits 1,300 characters ("how LLMs work" → "why the model forgets the start of a long chat"). Name the reader from the Voice section (devs who never used it, or people outside the field); unclear → ask.
2. Check every technical claim before drafting against the primary source: official docs, the spec, the source code or the paper. A version-dependent fact names the version (Unity 6, UE 5.6, the model name). A claim no source backs gets cut or asked about, never written from memory. Training data goes stale fastest on exactly these topics.
3. Build it around one concrete example: a number, a code fragment of three lines or fewer, or a moment in a game or tool the reader knows. An analogy comes after the precise statement, never instead of it.
4. Shape: the hook, the mechanism in plain words, the example, what the reader can do with it today, the close. Myth Bust (#11), The Comparison (#12), Curiosity Gap (#19) and The Direct Value (#21) fit teaching best.
5. The user's own use beats theory. One line on where they used it ("we hit this in our VR build") does more than a paragraph of background, and it comes from 01-Perfiles/Perfil Maestro.md or the user, never invented.
6. Four or more steps or parts are a carousel, not a post: hand over to /linkedin-carousel.

The receipt adds a sources: line with what each claim was checked against. The links go in the first comment, and each one is the primary page (the docs, arXiv, the repo), never a mirror or a summary site.

Phase 2: Hooks

Read references/hooks.json: 21 formulas, each with a template, a filled example, what it is for and how it usually gets ruined. Pick the three that fit this idea (three different formulas, not three versions of one), show them as numbered lines, and say in one sentence which you would ship and why.

Line 1 has to stand alone: the feed shows about two lines before "see more".

Phase 3: Draft

Line 1    the hook, alone
Line 2    the payoff of line 1, not setup for line 3
Body      paragraphs of 1-3 lines, a blank line between each
The turn  one line that reframes what came before
Close     one specific question, or one instruction; never both

• Length: 900-1,300 characters is the working range; the hard limit is 3,000. Under 400 reads as a thought.
• One idea per post. Two ideas are two posts: say so.
• Numbers over adjectives. No number given: ask, or leave {{your number}} and flag it.
• No engagement bait ("Thoughts?", "Agree?", "¿Qué opinan?"). The closing question is one only this post could ask.
• Three hashtags at most, at the bottom, only real categories people follow.
• Links in the first comment, not the body. LinkedIn has not confirmed that links reduce reach, but it is common practice; tell the user in the receipt.
• Language: the one the user's audience reads (the Voice section says it).
• Never fabricate metrics, clients, revenue or outcomes under the user's name, not even as placeholders.

Phase 4: Humanize

Run the draft through /content-humanizer before showing it:

python .claude/skills/content-humanizer/scripts/humanize.py draft.txt -o clean.txt --report
python .claude/skills/content-humanizer/scripts/detect.py draft.txt clean.txt

Rewrite the flagged lines by hand and re-run until PASS, or show the user the score and what is left.

Phase 5: Hand over

Print the post in a fenced block exactly as it should be pasted, then:

POST READY
hook:      #17 Time Anchor
length:    1,140 characters
humanizer: 6 artefacts stripped, human score 84 PASS
sources:   (explain mode) Unity Manual "Time and frame rate management"
post at:   Tuesday 8:15 (from the plan)
link:      goes in the first comment

Reply "yes" to log it, or tell me what to change.

On "yes", ask "May I append it to 01-Perfiles/LinkedIn Log.md?" and log the date, the hook and the first line, so the next plan does not repeat a theme.

Protocol

• Never publish, schedule or open LinkedIn in a browser. The User Agreement (section 8.2) forbids bots and automated methods that create, comment on or share posts, and an account flagged for it gets restricted.
• Credit: adapted from Jake Schincariol's linkedin-agent-skill (MIT).

Next steps

• /linkedin-reply once comments arrive (the first hour matters most).
• /linkedin-comment for the engagement round before posting.
• /linkedin-carousel for a topic with steps.
• /personal-brand to plan the next week.