Sharing agents lets you publish a versioned agent identity that others can install against their own services — with an explicit consent step before anything runs.

The three pieces

1. Published agent identity — a shareable, versioned declaration of what an agent is and what it needs. Locked at publication — the declared scope cannot change without a new version.

2. The consent screen — the moment of explicit, scoped approval. The installing user sees exactly what the agent is requesting before anything is granted. They can approve or decline.

3. The identity stamp — a verifiable badge the agent carries that anyone can inspect. Links to live data: current status, declared scope, install count.

The trust model

The agent author publishes a declaration. An identity provider (Gazebo) hosts the consent flow. The installing user grants access from their own connected services — the agent author never sees what they've connected, never touches their credentials, and cannot expand access without the user re-consenting.

This is the same trust model OAuth introduced for web apps — but applied to AI agents.

Where to start

• Publishing an agent — how to take an existing agent and make it shareable
• The consent screen — what the installing user sees
• Installing a shared agent — how the install flow works from the consumer's side
• The identity stamp — the verification badge and what it means