.skill

Skills hub

An AI skill is a new file format (.skill) launched by Anthropic, and it is now becoming the new open standard of the industry. This file format will change the way you work with AI. Try it: it is simple and needs no technical training, you just upload the file to a chat and write one word, Run

After free registration, the package of files with the skill comes to your email

How to use skills

  1. 1 · Email

    Registration by email

  2. 2 · One file

    The skill comes by email as one file with the extension .skill

  3. 3 · Chat

    You upload the file to a chat with ChatGPT or Claude and write one word: Run

In Gemini the file is uploaded once in Settings > Skills, and Gemini runs the skills that contain instructions without scripts

How a skill works

A skill is a file with the extension .skill that contains a technical package of AI instructions and deterministic scripts, optimised for precise work with files and with the abilities of the AI, and this gives a more solid, stable and predictable professional result

The format works by progressive disclosure: the assistant keeps only a short description of each skill in its context, and when a request matches it, loads the full instructions and runs the bundled scripts in its own code environment. The steps that need an exact result are executed as code, and the language model handles the steps that need judgement

duo-ready.skill

  • SKILL.mdinstructions for the assistant
  • scripts/code for the steps that need an exact result
  • references/knowledge that the assistant loads when it is needed

Featured

1 / 3

All skills

Supported assistants

  • ChatGPT

    ChatGPT accepts a .skill file in a chat and starts it after the word Run, and it also reads skills inside plugins, in the web app and in the desktop app

  • Claude

    Claude reads .skill files directly, in the web app, the desktop app, the mobile app and Claude Code

  • Gemini

    Gemini accepts a .skill file in Settings > Skills, in the web app, the Mac app and the mobile app, and replaces its Gems with skills in November 2026

Why it is safe

Every skill is plain text, so the instructions and the code of every script can be read before use, and a skill has no access of its own because it works only inside the permissions of the assistant. Every skill also passes the checks below before it is published, and the report is public

VirusTotalEvery .skill file is scanned by the antivirus engines of VirusTotal, a service of Google

OWASP Top 10 for LLM ApplicationsThe instructions are reviewed by hand against the ten main risks of AI applications, and prompt injection is the first of them

Semgrep · GitHub CodeQLThe code of every script goes through static analysis

Sigstore · SHA-256Every file is signed and has a public checksum, so a changed file can be detected

MITRE ATLASThe threat model of every skill is mapped to the known attack techniques against AI systems

NIST AI Risk Management FrameworkThe risk review follows the four functions of the framework: govern, map, measure and manage

European CommissionThe author is listed in the expert database of the European Commission

GDPR · BerlinSkills hub receives one thing from a user, the email address, and files, chats and results stay in the assistant