Approval Dashboard¶

Rampart includes an embedded web dashboard for managing approval-gated decisions. View pending approvals, approve or deny them, and review decision history — all from your browser.

Accessing the Dashboard¶

The dashboard is available when running rampart serve:

# Start the server
rampart serve --config ~/.rampart/policies/my-policy.yaml --port 9090

# Dashboard is at:
# http://localhost:9090/dashboard/

Authentication¶

The dashboard HTML is served without authentication — it contains no secrets. All data is fetched from the /v1/approvals API, which requires a Bearer token.

An interactive rampart serve can display the bootstrap token. Background and redirected service output omit the full token. Retrieve it in your own terminal with rampart token, then enter it in the dashboard. Keep it out of shared logs, screenshots, and agent prompts.

The browser stores it in localStorage for that origin and sends it to the configured Rampart service to authenticate requests.

Admin Scope vs. Eval Scope¶

Rampart credentials carry explicit scopes:

• Bootstrap admin token — persisted in ~/.rampart/token. It carries admin scope and can access all dashboard APIs.
• Named tokens — created with rampart token create <name>. They can carry eval scope, admin scope, or both, plus optional policy profiles.

Only a credential carrying admin scope can read or resolve dashboard approvals. Eval-only credentials are rejected by the approval and other admin APIs; they cannot turn an agent's own request into operator authorization.

Features¶

• Pending approvals: See all approval-gated decisions waiting for human input
• Approve / Deny: Resolve only the exact pending approvals shown in the confirmation
• Explicit run grants: Separately authorize future calls for the exact agent/session/run and credential-owner scope for the duration shown in the confirmation
• History: View past decisions with timestamps, agents, commands, and who resolved them
• Auto-refresh: Dashboard polls for new approvals automatically

Security Model¶

Signed URLs: When webhooks fire for approval-gated decisions, the notification includes a self-authenticating signed URL. Recipients can approve/deny by clicking the link without needing the Bearer token.

Network Access¶

By default, rampart serve binds to 127.0.0.1 (localhost only). Keep that binding for local use. Non-loopback bindings require TLS or a trusted HTTPS reverse proxy; an address alone does not establish a protected remote service. See the CLI reference for --tls-cert, --tls-key, and --tls-auto.

The dashboard token grants approval authority. Restrict access to the service and its credentials. Signed approval links likewise carry authority for the associated request and must stay private to intended reviewers.

API Reference¶

For authenticated request and response formats, see the canonical approval API reference.

Integration with Hooks¶

The dashboard resolves requests owned by Rampart's external approval queue, including Codex and Cursor hook requests. The waiting integration consumes the result; the dashboard itself does not execute the tool.

Claude Code's ordinary ask returns permissionDecision: "ask" and Claude owns the native prompt and resume. With ask.audit: true, the hook can mirror pending review data to the service and correlate later host events on a best-effort basis. Resolving that mirrored entry does not resume or cancel the native prompt. Claude's explicit ask.headless_only path instead waits on the external queue; see Native Ask.

OpenClaw native plugin approvals are owned by OpenClaw and do not create a second Rampart pending request. Consult approval paths and limits for other integrations.