On a run: permission mode and approvals
Choose the mode under Approvals in the composer with the first message. It stays fixed for that conversation.
A held action waits for a named person, and Any one approver settles it. What waits, where people decide and how long a run waits: Approvals.
Approval rules, who decides and how a decision applies
Approval rules, who decides and how a decision applies
Three rules make an action wait for a named person:
WHO APPROVES names people or groups. Any one of them settles a request. Self approve adds the person who starts the run.Gated commands are case-insensitive text fragments, such as
rm -rf or terraform apply. They wait in either permission mode, for anyone, and on schedules. Each takes an optional exception (unless), a note, approvers and notification channels. With no approvers, xpander asks the agent’s owner, and a night run parks instead of proceeding. Imported destructive scripts are gated by default.The held action shows a card with a sentence about the action (never its raw arguments), who started the run and who may decide. In Manually approve it is headed bash (gated mode).- A gated command or a Manually approve hold goes to the people or groups named on the rule, or to the agent’s owner if none.
- A marked skill action goes to the people or groups named under WHO APPROVES, plus the person who started the run when Self approve is on.
- When a hold has nobody to go to, the action does not run and the run continues. That happens to a marked skill action with nobody under WHO APPROVES and no run starter to self-approve, and to a gated command or Manually approve hold with no approvers on an agent whose owner has left.
- An approval can cover this exact call, the rest of the run, this kind of call for a few minutes, or until revoked.
- A decline answers only this attempt. The requester can withdraw a pending request.
- If nobody decides within the hold window (4 hours by default), a gated command does not run and the run continues. A later decision can still approve it. A marked skill action expires and the run resumes with “Not answered in time.”
GET /v1/tasks/{id} shows status: executing while one waits. Question cards are answered with POST /v1/agents/{agent_id}/conversations/{conversation_id}/answers (Submit Question Answers).
An approval card in the conversation.
On an agent: access and sharing
The person who creates an agent is its owner. Agent settings shows an Owner chip with their name. Ownership cannot be transferred.ACCESS is Org-wide by default: everyone in the organization gets the agent with full access, including editing. Manage access narrows it to Private (only you) or Group (people and groups you pick, then Apply). A restricted agent is invisible to others, and a link to it reads “not found”.
People with View or Use cannot change the agent: the server refuses their edits to its settings, card, shared workspace files and schedules. A Use grantee’s pick of CLI, model and approval mode in the composer applies to their own conversations only. They can still write files in their own conversation’s folder.
Group grants follow group membership. Whether a person may create, duplicate, delete or share agents at all is a user-group switch.

Manage access on an agent.

Manage access with Group selected: the people-and-groups picker lists your groups and members.

Choosing the permission to give: View, Use or Edit, then Apply.
ACCESS:
A hold that would go to the owner has nobody to answer it after they leave, so name an approver group on each rule.
On a conversation
Only the author may rename, delete or share a conversation. You can read it when it is yours, when you own the agent, or when the agent is org-wide. Share conversation opens it to your organization, named people or groups, or a public link. Allow replies lets a viewer continue in their own copy. Live surfaces follow the share.On a skill
A skill’s Shared with setting (Private, People, One agent or Organization) decides who can see and attach it. See Skill visibility.Whose credential a shared agent uses
Each skill is bound to a connection (an organization account or a personal one) and an auth mode:
A Vault secret can raise a binding’s mode, never lower it. Each call records the mode and whose credential was used, in the task and as a
use_secret audit event. MCP connectors also choose whether a person’s token is shared across their agents, and whether callers without their own sign-in may run.
Signed-in user (SSO) on an MCP connector
Signed-in user (SSO) on an MCP connector
Where single sign-on is set up, an MCP connector can use Signed-in user (SSO): the server gets the person’s SSO token as
Authorization: Bearer, with no separate sign-in.Runs from xpander Chat or Omni over MCP use the asker’s token. Schedules, webhooks, API calls and messages from unlinked accounts use the agent’s default user. A person without a valid SSO sign-in gets a note to sign in. xpander never opens OAuth for these servers. If a non-admin changes the address of a server that sends the raw token, it switches to Exchange for an audience.
Roles and user groups
A Member uses shared agents and creates agents where their group allows. An Admin manages everything under Settings and invites members. Platform > Organization > Members & groups has per-group switches for skills, agents, each agent-settings section, each Settings tab, workspace file actions and channels. A feature is hidden only when every group a person is in disables it. Groups can follow your identity provider.Next steps
Approve an agent's actions
What waits and who decides.
Connect accounts and credentials
How skills get credentials.

