Mos Geo Skills — AI skill for Claude Code

mos-geo-skills

Claude Code skills for GEO work — Generative Engine Optimisation. Built for The Vibe Marketing Lab.

GEO is the messy new corner of search where the "engine" is an LLM rather than a results page. A lot of what gets sold as GEO is guesswork. The skills in this pack are the opposite: each one ships with the evidence behind it, including the bits that say "this does less than people claim".

What's in here

Skill	What it does
mos-geo-llm-buttons/	Generates a row of "ask an AI about this page" share buttons for an article — ChatGPT, Claude, Perplexity, Grok, Google AI Mode — each opening with a pre-filled summarise-this-page prompt.

More skills get added as flat folders at the top level of this repo. One folder per skill, each with its own `SKILL.md`. Add the folder, re-run `setup.sh`, done.

`_shared/`

Cross-skill reference material. Not a skill, never linked into `~/.claude/skills/`, but every skill in the pack reads from it.

• platform-endpoints.json — the verified deep-link contract for each AI assistant (endpoint, query param, whether it auto-submits, whether it needs a login). Every entry was tested live, not copied off a blog. Treat this as the single source of truth and re-verify before any client build.
• geo-evidence.md — the honest evidence base for LLM share buttons. What the one controlled A/B test actually found, what the tactic does not do, and where the line sits between a UX feature and something Microsoft classifies as an attack. Read this before you pitch the tactic to anyone.
• logo-policy.md — per-provider trademark position on putting AI company logos on a button. Short version: wordmarks by default, logos opt-in.

Install

Skills live in `~/.claude/skills/`. This repo keeps them under version control and junctions them into place, so you edit here and Claude Code picks the change up straight away.

git clone  ~/Desktop/mos-geo-skills
cd ~/Desktop/mos-geo-skills
bash set