Agent permission boundary inspector

What an agent may read, write, call, deploy, or message externally — and under what scope. The denied / scoped / allowed colour language matches the capability matrix so a buyer learns one vocabulary across the family.

Included in the single kansho full pack. Components are not offered as separate installs.

Sandbox runner — restricted

Sandbox runner

Permission boundary

Read
fs:/workspacerepo:read-only
Allowed
Write
fs:/workspace/tmp
Scoped
Call
tool:shelltool:python
Allowed
Deploy
Denied
Message external
Denied

Plan & dispatch — broad

Plan & dispatch

Permission boundary

Read
fleet:*
Allowed
Write
tasks:*audit:write
Allowed
Call
agent:*
Allowed
Deploy
deploy:staging
Scoped
Message external
slack:#ops
Scoped

Usage

Guidance

List every verb the agent has — read, write, call, deploy, message-external.
Use the allowed / scoped / denied vocabulary shared with the capability matrix.
Show scope so a scoped grant reads as a boundary, not a blank cheque.

Don't

Rely on grant colour alone — each grant carries an icon and text label.
Invent verbs outside the shared permission vocabulary.

When to use: Reviewing exactly what an agent may do, and under what scope.
When not to: Editing permissions — this is a read-only inspector.

Accessibility

Screen reader

Announces

Each grant renders an icon plus an allowed / scoped / denied label, so the decision is never colour-only.

Not supported

The inspector is read-only — it exposes no interactive controls.

Props

API reference

Prop	Type	Description
agentNamerequired	string	Name of the agent whose boundaries are shown.
boundariesrequired	PermissionBoundary[]	Per-verb grants: verb, grant tone, optional scope list.

Component documentation

Permission Boundary Inspector

components/agent-setup/permission-boundary-inspector.tsx

Purpose

Shows an agent's read, write, call, deploy, and external-message grants with scopes.

Appropriate use

Use for a read-only permission review before assigning sensitive work.

Example

components/agent-setup/permission-boundary-inspector.tsx has a representative live example in the "Sandbox runner — restricted" section on /components/permission-boundary-inspector.

States

Verbs are read, write, call, deploy, and message-external; grants use shared permission tones and scopes are optional.

API and props

PermissionBoundaryInspector accepts agentName, boundaries, and className.

Dependencies

Direct imports are React, Lucide icons, AgentDot, ScopeChip, PermissionLegend, setup-state metadata, and cn.

Accessibility

Boundaries form a semantic list, and verb, grant, and scope are visible text so icons and colour remain supplemental.

Limitations

It is read-only and does not calculate effective permissions, resolve inheritance, edit grants, or explain policy sources.

Source

Source: components/agent-setup/permission-boundary-inspector.tsx.

Full pack

components/agent-setup/permission-boundary-inspector.tsx is documented at /components/permission-boundary-inspector and installs only through the single Kansho full pack; no individual component install is offered.