Jakeschincariol/linkedin-agent-skill

Eleven free Claude skills that run a LinkedIn account: posts off 21 hook formulas, comments, replies, profile score, weekly plan, and a humanizer that strips the AI fingerprint and scores the draft before it goes out.

What it is

LinkedIn‑agent‑skill is a collection of Claude “skills” (prompt‑templates plus a couple of tiny Python utilities) that help you write and manage LinkedIn content without ever sending anything to LinkedIn automatically. The repo ships eleven commands such as /li‑post, /li‑comment, /li‑profile, etc. Each command generates a draft (post, comment, reply, profile rewrite, weekly plan, carousel slides, DM sequence, inbox triage, audit, …) and then runs a humanizer that cleans invisible characters, normalises typography, replaces over‑used corporate‑speak, and scores the draft on five heuristics (burstiness, specificity, slop density, fingerprint, voice). The final output is a copy‑ready block that you manually paste into LinkedIn – the tool never uses the LinkedIn API.

How to get it running

  1. Claude installation – copy the li‑* skill folders into ~/.claude/skills/ (or the repo‑local .claude/skills/). In Claude you can also paste the repo URL to install the skill directly.
  2. Optional plugin – the repo can be added via Claude’s plugin marketplace (/plugin marketplace add …).
  3. Humanizer scripts – two Python 3 scripts (humanize.py and detect.py) live in skills/li‑human/. They have no external dependencies; just run them on a draft file to clean it and see the five‑check scores.
  4. Voice profile – fill templates/voice.md (or ~/.claude/linkedin/voice.md) with examples of your own writing so the generated drafts match your personal tone.

Core features (the eleven commands)

Command Purpose
/li‑post Turn a single idea into three hook options (from 21 formulas) and a full draft, then human‑filter it.
/li‑comment Generate a context‑aware comment (nine styles) for someone else’s post – never a generic “Great post!”.
/li‑reply Organise and answer comments on your own post, classifying them (lead, substance, peer, support, noise) and drafting replies in that order.
/li‑profile Score your LinkedIn profile on a 12‑point rubric (out of 100) and produce a rewrite that fixes the lowest‑scoring items first.
/li‑plan Produce a weekly posting plan, including timing and a list of ten people to engage with, saved as ~/.claude/linkedin/plan.md.
/li‑human The humanizer: removes zero‑width characters, normalises punctuation, replaces “slop” buzzwords, and runs the five‑check detector.
/li‑carousel Build a document‑style carousel post: slide‑by‑slide copy, an eye‑catching cover, and a ready‑to‑upload PDF.
/li‑repurpose Convert a video, newsletter, or transcript into a week’s worth of standalone LinkedIn posts.
/li‑dm Draft a 200‑character connection note, an opening message, and two follow‑up messages for outreach.
/li‑inbox Triage incoming LinkedIn messages into categories (lead, recruiter, peer, ask, spam) and highlight language that may give away automation.
/li‑audit Analyse already‑published posts, ranking them by engagement rate and reach (not just impressions).

The “humanizer” in detail

  • Cleaning passes – humanize.py strips invisible Unicode characters, converts em‑dashes to commas, curly quotes to straight quotes, etc.
  • Lexicon replacement – a JSON file (slop.json) lists 113 over‑used corporate terms and their plain‑English alternatives; you can edit it.
  • Detection – detect.py scores a draft on five heuristics (burstiness, specificity, slop density, fingerprint, voice). The final human score is a weighted blend (60 % average, 40 % worst‑case check) to surface drafts that still look AI‑generated.

Limitations & legal notes

  • No automatic posting – the tool deliberately stops short of using LinkedIn’s API (which is unavailable for personal profiles) and instead gives you a copy‑ready block to paste manually. This avoids violating LinkedIn’s User Agreement.
  • Local heuristics only – the five checks are simple rule‑based metrics, not integrations with commercial AI‑detector services. Scores are indicative, not guarantees of “undetectability”.
  • Python scripts are dependency‑free but require a local Python 3 interpreter.
  • Voice profile required – without a filled voice.md the generated text falls back to a generic style.

Who built it

Created by Jake Schincariol (see opusjake.ai). The full write‑up is linked in the README.

License

MIT – you’re free to fork, modify, and redistribute.

Related

  • Project
    clawdotnet/openclaw.net

    OpenClaw.NET is a .NET‑native AI‑agent runtime and gateway. It provides a NativeAOT‑compatible executable, a web‑based chat/admin UI, OpenAI‑compatible APIs, 80+ built‑in tool adapters, multi‑channel support (Slack, Telegram, etc.), and first‑class integration with cloud and local LLM providers (OpenAI, Claude, Gemini, DeepSeek, Ollama, embedded Gemma‑4). Optional governance features let you audit and approve high‑risk tool actions. The project is aimed at developers who want a self‑hosted, observable AI‑agent platform that runs on Windows, macOS, or Linux.

  • Project
    Autodesk/XLB

    XLB is a Python library for differentiable Lattice Boltzmann fluid simulations. It runs on CPUs, single‑GPU (Warp), multi‑GPU (Neon), and TPU (JAX) hardware, offers grid‑refinement, many LBM models and boundary conditions, and integrates with JAX‑based ML tools for gradient‑based optimisation and physics‑informed learning.

  • Project
    ollama4j/ollama4j

    Ollama4j is a Java library that wraps the Ollama local LLM server, providing text generation, chat, tool calling, embeddings, image inputs, model management, and Prometheus metrics. It is distributed via Maven Central/Gradle and integrates with any Java project.

  • Project
    Optima-CityU/LLM4AD_Next

    LLM4AD Next is a Python toolkit that turns a natural‑language problem description into a full, runnable evolutionary‑algorithm search project. An interactive LLM consultant generates the code skeleton, configuration, and evaluator, then built‑in evolutionary methods (IslandGA, Diverse Island GA, EoH, etc.) automatically evolve the generated algorithm. It also offers an advisor to suggest evolution‑ready code blocks, long‑term memory support, Docker images, and an online demo.