> ## Documentation Index
> Fetch the complete documentation index at: https://docs.runr5e.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Identity and Delegation

> Every action is attributable. Authority flows down, never up.

## Four Identity Layers

| Layer             | What                                     | Purpose                             |
| ----------------- | ---------------------------------------- | ----------------------------------- |
| **Tenant**        | Organizational boundary                  | All resources scoped to a tenant    |
| **Human (Owner)** | The person ultimately accountable        | Every chain terminates at a human   |
| **Agent**         | Durable template (model, harness, tools) | Defines what kind of worker this is |
| **AgentSession**  | Live execution context                   | The unit of accountability          |

## Dual Identity

Every event, artifact, and API call carries both the session identity AND the owner identity:

* "What did Jane's agents do last week?" → query by owner
* "What did the researcher agent do?" → query by agent
* "What happened in this session?" → query by session

## Delegation

Authority flows downward through the `parentSessionRef` chain:

* A child's authority is a **subset** of its parent's
* A child's timeout cannot exceed its parent's remaining time
* The child inherits the parent's owner (accountability doesn't change)
* Delegation goes through the **admission pipeline** — no side-channel creation

## Credential Isolation

Agents never hold secrets. They hold **leases** — time-bounded, session-scoped, revocable references that the runtime resolves at the point of use. The credential value never appears in events, artifacts, or session records.