Linkedin Setup

Setup

Paths in this file are relative to the bundle root, the directory holding the router SKILL.md, not to this skill's own directory. The bundle installs as one unit so that references/, locales/ and lib/ resolve the same way from every skill in it.

This is the entry point, and for anyone who did not write this engine it is the actual product. Without it, a published bundle is a folder of prompts.

It takes about twenty minutes and it ends on a written post, not on a file.

What it produces

One file, written out in this conversation for the person to save. Not a directory, and not a file this skill creates: the floor writes nothing to disk. What comes out is text on screen, with a name suggested for it, and the person keeps it wherever they keep things.

Its format is references/material.md, and it has six sections: Status, Profile, Voice, Pillars, Ideas, and a Ledger that starts empty.

Their published posts stay outside it. They are pasted into this conversation at step 4, read once for the Voice section, and stay wherever the person already keeps them. A hundred post bodies inside an attached file is the one thing that does not scale.

A tier that can write files may offer to save it. That is a convenience, and it never changes what this skill promises.

Step 0. The language

Do not ask for a path, and do not assume one. There is no directory to choose. A tier holding a filesystem may suggest where to put the file once it exists; this skill produces the text.

Settle the three language axes from SKILL.md, in one exchange:

• Interview language. The one they want to be talked to in.
• Output language. Defaults to the same. Ask explicitly, because plenty of people work in one language and publish in another, and this is the single most common reason a bundle feels foreign.

If no pack exists for the interview language, say so plainly, fall back to locales/en, and offer the pack template as a contribution. Do not machine translate a pack on the fly and present it as one.

Step 1. The basics, fast

Fill the top of the Profile section with short answers: what they sell, who they write to, what this channel is for. Two minutes. These are facts, not reflection, and dwelling on them is how an onboarding loses people before the part that matters.

Two fields deserve their own attention because nothing else in a profile captures them and their absence produces tutorials:

• What they can prove. One bullet per fact, each with a source. A number without a source is a memory, not a fact. Push for the unflattering ones: they are the most publishable material anyone has.
• Names that must never be cited. Explicit list. Everything not on it and already public is free to use.

Step 2. The interview

Ten intents, in references/interview-intents.md, set A. Wording in locales/<interview_language>/interview.md.

Every question announces what the answer becomes. Not afterwards, in the question itself. A person who knows they are feeding their public positions answers differently from a person filling a field called "beliefs".

Match the affordance to the question, per the three shapes in the intents file. This is the part that decides whether a stranger finishes.

Ticking a box costs three seconds and yields a usable sentence. "What is your deepest conviction?" in an empty field costs ten minutes and usually yields a platitude. Constraint produces more than a blank page.

Build the propositions from what this person has already said, never from a template of what a good answer looks like. Otherwise they end up agreeing to someone else's opinion, and every post afterwards is written on top of it.

Step 3. The pillars

Three. Assembled from the interview, not chosen from a list.

Write them as postures, not as subjects. A subject says what a post is about. A posture says what it does to the person reading. This is the difference between a ratio that can be argued about and a ratio that is arbitrary.

Each pillar carries:

• A definition, one or two sentences, specific to this person's market.
• An effect label: what the reader leaves with. Authority, trust, or connection.
• A verifiability requirement, if the person has one. A pillar built on numbers dies quietly if it is allowed to run on opinion.

Then the ratio, out of the monthly count from the cadence intent. Derive it from the business objective and say the derivation out loud, so it can be disagreed with:

Write the Pillars section with the counter rule: at the start of every session, count the pillars of the last N ledger entries, and if one is more than two behind, offer it first.

Step 4. Voice

Ask them to paste three to five published posts into this conversation, up to six. Their own, not ones they admire. Do not ask for a folder and do not ask for links: what cannot be read here cannot be quoted from, and a link is a promise this skill cannot keep.

Read them and write the Voice section: the traits actually present, quoted. Form of address, hook shape, paragraph rhythm, how they close, real length.

Every trait prints the sentence it was read from, directly under it. No code checks this any more, so the quote is what makes a guess visible to the person instead of to nobody:

**Opens on a number that sounds wrong.** Not on a claim, on a figure the
reader has to reconcile.

> "Eleven hours. That is the median time I spend building a first board
> pack, and most of it is not spent on the numbers."

A trait without a quote under it is a guess and does not go in. Trimming those quotes as redundant removes the enforcement rather than tidying it.

If there are fewer than five posts, open the Voice section with a banner saying it is provisional and built on N posts, and instruct every skill reading it to defer to the hard style rules instead of to the observed traits. The banner comes off when the corpus is real, not when it feels awkward.

If there are no posts at all, say so, write the Voice section with the hard rules only, and note that it gets rewritten after five to ten published posts.

The hard rules are the ones marked hard in locales/<lang>/style.md, which is why this skill carries that pack: a Voice section built on a short corpus, or on none, is written out of it rather than out of traits nobody could read.

Never build a voice from a scraped corpus of high performing posts. It produces the average of a niche, which is the opposite of a voice. Studying five people they actually respect, by hand, for structures and never for turns of phrase, is a different activity and is fine.

Step 5. The idea bank

Nine ideas, three per pillar. Each one is an angle, which already contains a position, not a subject.

Each carries:

• A title.
• A funnel label: VISIBILITY, TRUST or ACTION. See references/formats.md.
• Two lines of pitch.
• The material that already exists for it, named. An angle with no material behind it is a wish, and it will be the one that stalls a session.

The dosage of the labels follows the business objective, not taste. Someone building authority runs heavy on VISIBILITY. Someone with open slots inverts it. Say the dosage out loud when you write the bank.

Two ergonomic rules worth copying:

• Regenerate one slot, not the whole grid. Nobody should have to choose between keeping everything and throwing everything away.
• Go straight from an idea into the interview. No copying a title into another screen. The gap between having an idea and starting to write is where sessions die.

Step 6. Publishing

At the floor there is nothing to settle here. The post comes out as text and the person posts it themselves, which is what copy always meant.

Ask only if the host can run commands. Then: postiz needs an integration id, and the id must be checked against the channel list first, because a personal profile and a company page are two lines in a config file and two very different things in a feed. command runs their own binary. See lib/publish.py. A tier adds convenience here and never adds a promise, so do not describe scheduling to somebody whose host cannot do it.

Step 7. Do not end on a file

Hand straight over to linkedin-post and write one. Right now, with an idea from the bank that was just built.

A setup that ends on a finished material file has produced a file. A setup that ends on a published post has produced a habit. The difference between those two outcomes is the difference between this working and not. The floor makes this truer rather than harder: there is no folder to admire at the end, only text the person is holding.

Then set the next session, with a date and an idea attached to it.

And only then, write the ## Status block at the top of the material for real: filled: yes, source: interview, updated set to today, and the two language codes settled in step 0. Five lines, and they are the whole seam between the engine and this person. Every consumer reads them before it reads anything else, so they are the last thing this skill writes, not the first.

Rules

• Nothing in this skill writes into this repository. The profile lives on the person's machine, at the path they chose.
• Never fill a field on someone's behalf and move on. Proposing three options is help. Picking one for them and continuing is authorship.
• Do not translate a language pack on the fly. Fall back, announce the fallback, and offer the template.
• filled: yes only when the sections are real. The flag is what every other skill trusts. Setting it early makes every downstream skill lie confidently.