# Shared skills for teams — Orlo

Markdown version of https://useorlo.com/skills.html, generated 2026-10-14 from the page itself. Illustrations and interactive examples are left out.

**One published version of your team’s instructions, wherever they run.**

A skill is a reusable set of instructions with an owner and a version. Draft it, test it in a task, publish it, and every release, agent, and AI tool that uses it runs the exact version that passed.

[Read the skills documentation ↗](https://docs.useorlo.com/)

## What a skill is.

### Every save is a new version that never changes.

Instructions, up to 24 supporting text files, and a content hash. Editing creates the next version and leaves earlier ones exactly as they were.

### Every skill has an owner.

A person or a team is responsible for what it says, and publication records who published it.

### A draft is tested before it is published.

Select a draft version in a task, run the comparison, read the result, then publish that exact version. A release that still points at a draft is refused.

## Where a published version runs.

### In task releases.

A task pins up to eight skill versions in a chosen order. Their instructions and reference text go to the model ahead of the task’s own instructions, in comparisons and in production alike. Scripts are never run and never pasted into a prompt.

### In Claude and ChatGPT.

A read-only connector gives each employee the organization’s published skills under their own sign-in. A connection is scoped to one organization, can be revoked by its owner, and each retrieval is recorded.

### In Claude Code.

An installer places an exact published version in a project’s skills folder, verifies its content hash, and refuses to overwrite a skill that is already there. Upgrading or rolling back is explicit.

### In your own agents.

The agent SDK loads exact versions, checks their hashes, and records which versions a run declared, so the run’s record shows what the agent was following.

## What stays on record.

### Who uses a version.

For any version: the tasks that pin it, the comparisons that tested it, and the agent runs that declared it.

### Who published it, and who retrieved it.

Publication and each employee retrieval through the connector appear on the evidence page, beside the task’s other records.

## Bring the skills you already have.

### Import a file and review it before saving.

A Markdown file, a .skill or .zip package with its supporting files, or an Orlo export. The console reads it locally, shows what it found, and saves a draft only when you say so.

### What to know before you rely on it.

A published version cannot be withdrawn yet; publish a successor and move tasks to it. Installing a skill in a tool does not report that tool’s runs to Orlo; the record covers retrievals and the runs your agents declare.

## Decide what an agent following those skills may do.

## More

Orlo is a fineSample product. Documentation: https://docs.useorlo.com/

Other pages as Markdown: [Overview](https://useorlo.com/index.md), [Evaluation](https://useorlo.com/evaluation.md), [Agent permissions](https://useorlo.com/agents.md), [In production](https://useorlo.com/deployment.md), [Model swap regression](https://useorlo.com/model-swap-regression.md), [Model catalog](https://useorlo.com/catalog.md). Index: https://useorlo.com/llms.txt