28 free, MIT-licensed skills that turn Claude Code, Cursor, or any skills-compatible agent into a working LinkedIn content assistant. Each skill is a distilled methodology your agent's own model runs locally — no API keys, no accounts, nothing phones home. Humans: the same tools live at liftli.ai/tools.
A skill is an installable instruction package — a SKILL.md your agent reads and executes with its own model. Installing one teaches your agent a content craft: the hook patterns that survive LinkedIn's "see more" fold, the 8 checks of a strong headline, the tells that make writing smell like AI. The knowhow is the same behind Liftli's 28 free web tools; skills are the version built for the terminal.
Install everything with the command above, or a single skill with --skill <name> — every card below carries its exact command.
Generates 8 scroll-stopping opening lines for a LinkedIn or X post, each built on a different proven pattern (contrarian claim, number + unexpected outcome, confession, mid-story drop) and sized to survive the ~210-character “see more” fold. The agent's own model does the writing using the distilled pattern library — no API key, no account.
Writes a complete LinkedIn post from a topic or raw material: hook above the fold, 1–2 sentence paragraphs, concrete specifics, one clear takeaway. Encodes the editorial structure that separates posts people read from posts people scroll past.
Rewrites a draft post for reach while keeping the author's facts and voice: fixes buried hooks, wall-of-text formatting, AI-sounding phrasing and weak endings, then explains every change it made. Never invents claims that weren't in the original.
Interviews the user about their actual expertise, then mines it for 15 specific post ideas across five angles: lessons & mistakes, contrarian takes, process, stories, and data. Each idea is one sharp sentence someone could immediately start writing from — no generic listicles.
Writes 25–40 word LinkedIn replies with actual substance — a concrete example, a respectful challenge, an extension, a micro-story, or a genuinely useful question. Never opens with empty praise and never includes a CTA.
Structures a LinkedIn carousel (a PDF document post) slide by slide: cover hook, one idea per slide in under 25 words, recap, soft CTA — 8–10 slides. Text before design, because a pretty carousel with mushy slides still fails.
Creates complete LinkedIn polls people actually answer: a question under 140 characters the reader feels qualified and eager to vote on, 3–4 chooseable options under 30 characters each, and an intro post that stakes a light opinion so comments have something to react to.
Repurposes a LinkedIn post for X (Twitter) properly instead of copy-pasting: a standalone post under 280 characters, a 4–6 post thread whose first post hooks alone, and a spicier quote-bait variant. Strips LinkedIn-isms and tightens to the core claim, because X rewards conversation.
Generates 6 defensible contrarian angles on a topic — each with the take itself, the steelman of the common advice, and the conditions under which the contrarian view wins. Built for honest disagreement, not shock-bait; helps the user pick the angle they can defend from lived experience.
Turns a developer's real work week — commit messages, PR titles, or a described week — into a build-in-public LinkedIn post with actual narrative tension (the bug that took three tries, the feature that got cut). In a git repo, the agent can read the log itself. Never a changelog dump.
Mines meeting notes or call transcripts for their strongest post-worthy insight — the customer sentence that reframes the problem, the objection everyone has, a decision and its reasoning — and writes it up fully anonymized, plus two runner-up angles.
Writes 7 LinkedIn headlines under 220 characters using different proven formulas (“I help X achieve Y through Z”, outcome + proof, role | niche | proof-point), front-loading the first ~65 characters that survive search truncation. Flags buzzwords and pipe-soup.
Scores an existing LinkedIn headline against 8 checks: length vs the 220 limit, the ~65-character search cut, buzzwords, pipe count, all-caps, audience signal, and the title-only trap — with a specific fix for every failed check and one rewritten example.
Writes a LinkedIn About section that reads like a landing page, not a bio: a hook in the first 3 lines (all that shows before “see more”), who you help and the outcome, proof with numbers, how you work, and exactly one call-to-action. First person, 150–250 words.
Walks the user through a 25-point profile audit in five groups — photo & banner, headline & URL, About, experience & skills, activity & social proof — with a weighted score (activity and About count double) and a prioritized fix list.
A GEO (generative engine optimization) audit for a person: 12 checks across Presence, Consistency and Citability that determine whether ChatGPT, Claude and Perplexity can find and cite the user for their expertise. Returns a tier (Invisible / Fragments / Citable) and concrete next actions.
Check — Pre-publish checks the agent runs in seconds
Check
linkedin-character-limits
LinkedIn Character Limits
A reference skill carrying every LinkedIn character limit for 2026 — posts (3,000), the ~210/~140 character “see more” folds, headline (220), About (2,600), comments (1,250), articles, messages, polls — plus the editorial rules of thumb, so the agent can length-check drafts in the terminal.
Computes exactly what survives LinkedIn's “see more” fold: shows the first ~210 characters (desktop) and ~140 (mobile) as readers will see them, flags cuts that land mid-word or mid-idea, and judges whether the amputated hook still creates pull.
Applies a 10-point pre-publish checklist to a draft: hook fits the fold, sentence length, wall-of-text paragraphs, total length, hashtag and emoji counts, buzzwords, ending strength, bare URLs. Framed honestly as editorial best practices, not algorithm claims.
Audits a draft for the tells that make readers smell AI — filler openers, “Not only… but also” constructions, enumerator adverbs, tell-vocabulary like “delve” and “seamless”, em-dash chains, uniform sentence rhythm — then fixes by adding what's missing (real numbers, real moments, actual quotes), not by paraphrasing the tells away.
A pre-publish second opinion on a spicy draft: a verdict (SHIP IT / SHIP WITH EDITS / RETHINK), the most uncharitable plausible reading, ranked risks, and the minimal edits that keep the edge without the backfire. The goal is braver posts, not neutered ones.
Converts text to Unicode bold, italic, bold-italic or monospace for LinkedIn using the Mathematical Alphanumeric Symbols mapping — with the honest caveats: screen readers spell it letter-by-letter and search can't index it, so use it for one or two emphasis moments, never whole posts.
Cleans text pasted from Google Docs or Word before it hits LinkedIn: strips zero-width characters, normalizes exotic spaces, collapses runs of blank lines, trims trailing whitespace, and reflows paragraphs to the 1–2 sentence rhythm the feed rewards.
Measure — Metrics, timing, costs and reference tables
Measure
linkedin-engagement-rate
LinkedIn Engagement Rate
Computes LinkedIn engagement rate both ways — by impressions (for your own posts) and by followers (for comparing accounts) — and interprets the number with clearly-hedged rules of thumb, small-sample caveats, and why comments outweigh reactions.
Projects follower growth with compounding math and milestone estimates, wrapped in the honest caveats: real growth comes in spikes and plateaus, followers is a lagging metric, and consistency changes the slope more than any tactic.
The honest answer on LinkedIn timing: there is no universal best time — published studies disagree because they average different audiences. Carries the commonly-reported windows plus a 4–6 week method to find the user's own best slots from first-hour impressions.
Prices the three ways to keep a LinkedIn presence alive: a human ghostwriter ($500–$3,500/month), DIY (the user's hourly rate × hours per post), and software. Helps the user compute their own numbers and says plainly when a human is the right call.
A reference skill with every LinkedIn image dimension for 2026 — profile photo, personal banner (1584×396), company pages, post images, article covers, carousel PDFs — labeled recommended vs minimum, plus the crop rules (the profile photo overlaps the banner's bottom-left on desktop).
Skills are stateless: your agent brings the model, the skill brings the craft. What they can't know is you — your writing voice, your strategy, your material, what worked last month. That's the Liftli MCP: it extracts your voice from your real posts, mines your voice notes, calls and GitHub activity for material, keeps persistent strategy memory, and publishes to LinkedIn, X and Substack behind a one-tap approval gate. Connector details for agents: liftli.ai/llms.txt. Free tier, no card.
Frequently asked questions
What are agent skills?
Skills are installable instruction packages for AI agents — a SKILL.md file carrying a distilled methodology that the agent's own model executes. Installing a Liftli skill teaches your agent how to do one content task well (write hooks, audit a profile, check timing) using the same knowhow behind the free tools at liftli.ai/tools. No API keys, no accounts, nothing phones home.
Which AI agents and clients support these skills?
Any client compatible with the open skills format: Claude Code, Cursor, and other agents that read SKILL.md packages. Install with npx skills add liftli-ai/skills. The skills are plain markdown, so even agents without a skills manager can be pointed at the GitHub repository and use them directly.
Are the skills really free? What's the catch?
MIT-licensed, free forever, and they run entirely on your agent's own model — Liftli never sees your content and pays nothing per use, so there is nothing to meter. They exist because they're the single-serving versions of what Liftli does as a full pipeline; if a skill earns a place in your workflow, you know where the rest lives.
How do skills differ from the Liftli MCP?
Skills are stateless methodology — your agent brings the model, the skill brings the craft. The Liftli MCP is the stateful product: it extracts your writing voice from your real posts, mines your voice notes, calls and GitHub activity for material, keeps persistent strategy memory, and publishes to LinkedIn, X and Substack behind a one-tap approval gate. Skills need no account; the MCP has a free tier at liftli.ai.
Do the skills work offline or in CI?
Yes — a skill is a markdown file on disk. Once installed it works wherever your agent works, including air-gapped or CI environments, because there are no network calls: the agent's own model does all the generation and analysis.
Can I use these skills for a human, in a browser?
Every skill has a web twin at liftli.ai/tools — 28 free browser tools with the same methodology, no login required. The skills page and the tools page cover the same ground for two audiences: agents install skills, humans use the web tools.
Your agent knows the craft. Liftli knows you.
Voice extraction, strategy memory, real material from your week, publishing behind your one-tap yes — connect the Liftli MCP.