Context
Identify the topic, source material, timing and the user's angle.
TVL AI Skills
LinkedIn Post Writing Skill
tvl-write-linkedin-post turns a paper, launch note, article or rough draft into a technical LinkedIn post. It analyzes the source with the CROFTAFC framework, drafts under LinkedIn Writing Protocol v3, then runs a deterministic validator and revises until every rule passes.
The framework
CROFTAFC: the pre-draft analysis
Before a single sentence is written, the skill answers eight questions. Each letter fixes a decision, so the draft starts from evidence and intent — not a blank page.
Role
Write as an experienced engineer or AI researcher addressing a technically curious peer.
Objective
Teach one useful idea through evidence, mechanism and a supported conclusion.
Facts
Extract measurable results, dates, technical claims, mechanisms, limitations and source links.
Tasks
Select the strongest evidence, build a logical paragraph sequence, draft, validate and revise failures.
Audience
Address professionals interested in AI and tech; stay accessible to informed non-specialists without losing substance.
Format
Plain paragraphs only, in the requested language and length (default 2,700–2,900 characters).
Constraints
Apply all Protocol v3 rules; user instructions override only length, language, topic or structure.
The rules
LinkedIn Writing Protocol v3
The protocol is the standard the draft is measured against. It fixes the numbers, the structure and the language.
2,700–2,900Target characters for the whole post, unless you ask for another length
150–250Preferred characters per paragraph, for a natural reading rhythm
≥ 10Words per sentence as a rule; avoid short punchy lines unless clarity needs one
| Area | Rule |
|---|---|
| Hook | Open with a measurable fact, research result, concrete technical observation, or a strong claim immediately backed by evidence. Never open with generic intro, background, a rhetorical question or empty context. |
| Structure | Order paragraphs so each one adds evidence, explains a mechanism, or derives a consequence. No repeated ideas or conclusions. |
| Formatting | Plain paragraphs only. No bullets, numbered lists, headings, hashtags, emojis, tables or Markdown links. Plain source URLs only when useful. |
| Punctuation | No em dashes; avoid hyphens where a natural rewrite works. |
| Voice | Explain as if to a technically curious colleague: simple language, professional tone, technical precision and a natural rhythm. No AI cadence or mechanical paragraph structure. |
| Evidence | Every sentence carries evidence, a mechanism, or a supported conclusion. Predictions must be backed by prior evidence in the post. |
Banned rhetorical structures
not X but Yless X more Yinstead of Xforget Xthe real story isX is deadX no longer matters
Banned editorial transitions
Another important trendLooking aheadThis matters becauseThe challenge isIt is worth notingIn conclusion
Banned marketing language
this changes everythinggame changermind blowinghugerevolutionarytransformative
Implementation
The nine-step workflow
Drafting is only half of it. The skill verifies facts first, then runs a script that fails the draft on any protocol breach and revises until it passes.
Read the sources
Extract results, mechanisms, numbers, limitations and business consequences from everything supplied.
Verify
Confirm research or current facts against primary sources: papers, repositories, announcements, benchmarks.
Choose one idea
Pick a single central technical idea and order the paragraphs to build it step by step.
Draft
Write in simple, professional, direct language following the protocol.
Validate deterministically
Run validate_post.py to check character count, paragraph and sentence length, and rule compliance.
Fix and rerun
Correct every reported error and rerun the validator until it passes clean.
Semantic checks
Judge what software cannot: no repeated ideas, every sentence earns its place, predictions supported, natural rhythm.
Rewrite and revalidate
Rewrite any semantic failure and run the whole check again.
Return the post
Return only the finished post, unless you asked for sources, analysis, alternatives or validation details.
The deterministic validator
A Python script enforces the mechanical rules so the model cannot talk its way past them.
$ python3 scripts/validate_post.py draft.txt
Custom length range:
$ python3 scripts/validate_post.py draft.txt --min-chars 1800 --max-chars 2200
It checks: character count in range, paragraph length, no needless short sentences, banned phrases and transitions, no em dashes or hashtags or emojis or lists or headings or tables, and that the opening starts from evidence.
Under the hood
What ships with the skill
references/protocol-v3.md
The complete LinkedIn Writing Protocol v3 — the full source reference for every rule above.
references/evaluation-cases.md
Testing and modification cases used to check the skill's behavior against expected output.
scripts/validate_post.py
The deterministic validator for character count, paragraph and sentence length, and protocol compliance.
Install and run
Add the LinkedIn skill
Use the public TVL Skills repository with the Agent Skills CLI.
Install only this skill
$ npx skills add danlex/tvl-skills --skill tvl-write-linkedin-post
Install all TVL skills
$ npx skills add danlex/tvl-skills
Turn a paper into a post
Use $tvl-write-linkedin-post to turn this paper into a LinkedIn post. Lead with the strongest measurable result, explain the mechanism, and keep it in the 2,700 to 2,900 character range.
Write from a product update
Use $tvl-write-linkedin-post to turn this product update into a LinkedIn post for a technical audience. Preserve the numbers and validate it against Protocol v3 before returning.
Shorten and revalidate
Use $tvl-write-linkedin-post to shorten this draft to 1,800 to 2,200 characters, keep every fact, and rerun the validator until it passes.