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

# Governance and safety

> How SuperEmber scopes identities, reviews actions, preserves human control, and keeps work auditable.

SuperEmber is designed for delegated work with explicit boundaries and human control.

## Core principles

* A conversation provides intent and context; it is not unlimited authorization.
* Personal, team, company, and executive identities have different access scopes.
* Delegation follows visible reporting relationships.
* Boundary-crossing tool calls are reviewed before execution.
* Risky or ambiguous exact actions pause rather than silently executing.
* Approvals are one-use decisions, not blanket permissions.
* Durable runs, action reviews, delegation records, and receipts preserve an audit trail.
* Secrets stay out of prompts, chat, screenshots, templates, and Emberbase.

## Identity and access boundaries

| Boundary          | Guardrail                                                                    |
| ----------------- | ---------------------------------------------------------------------------- |
| Personal Agent    | Only the owner can access private credentials, memory, and approvals.        |
| Team Agent        | Team membership and capability policy control invocation and management.     |
| Company AI        | Organization-wide general entry point and channel fallback.                  |
| Executive AI      | Protected coordination identity with stricter membership and mutation rules. |
| Cross-tenant work | Denied; an action cannot change tenant identity during review or execution.  |

Assigning a channel, connector, or reporting line never overrides these access checks.

## Review before execution

Automatic review evaluates the exact proposed tool call. It can approve, deny, or require a human. If the action changes after review, the approval no longer applies.

Delegated instructions receive stricter authorization treatment because another agent—not the human—authored the handoff.

See [Automatic review and escalations](/product/auto-review-and-escalations).

## Human decision points

Use a human decision for:

* external messages and customer-facing posts
* destructive or permission-changing actions
* financial or legal commitments
* public disclosure or credential-sensitive operations
* workflow steps where policy requires an accountable reviewer
* ambiguous actions that do not clearly match the user's intent

Personal connector approvals, delegated escalations, workflow waits, and channel-binding approvals are separate lifecycles. Use [Approvals, review, and inbox](/product/approvals-and-inbox) to find the right surface.

## Connector governance

Connect an application to the narrowest agent that needs it. Team connectors serve the Team Agent; personal connectors serve only the Personal Agent.

Disconnect a connector when the account, employee, team responsibility, or intended automation changes.

## Workflow governance

Workflows are validated before test runs and publishing. Published versions remain immutable for run history. Templates strip live bindings into placeholders so imported packages do not silently inherit credentials or destinations.

Never place credentials, tokens, customer secrets, or raw environment files inside workflow definitions, prompts, templates, docs, or memory.

## Emberbase safety

Use Emberbase for durable business context, decisions, procedures, and current facts. Keep personal context in Personal Emberbase and team context in the appropriate teamspace.

If a fact is wrong, correct the source page. Do not try to override durable knowledge with a later chat message.

## Public files and screenshots

Review files and screenshots before sharing them publicly. Authenticated product screenshots can expose names, channels, customer context, or credentials even when the original request seems harmless.