Use the Governance dashboard

Use the Governance dashboard when you want to put a USD spend limit on your agents, resume a scope that was auto-paused at its cap, clear the tool/delegation approvals that are blocking a run, or read the forensic audit log of what happened. It is the human-facing surface for Clawboo’s trust controls. The panel lives at the Governance nav view (GovernancePanel), backed by /api/governance/*. It has four sections, top to bottom: Budgets, Caps (enforced in code), Approval queue, and Audit log. This page documents what each does, the options, and how to verify. For the concepts behind these controls, the budget kill-switch, circuit breakers, and builder≠judge verification, see Governance. For the raw request/response shapes, see the /api/governance reference.

​

Prerequisites

• A running Clawboo dashboard (the panel polls /api/governance/budgets every 5 seconds and /api/governance/audit on each filter change).
• To see budgets do anything, you need agents that incur cost; spend is recorded by the executor loop as runs produce cost events.

​

The default posture

Clawboo ships track-and-warn: nothing pauses your agents out of the box. A budget is a row you opt into per scope, and each budget has a mode: The pause-at-cap is automatic in the executor; the UI only sets budgets, raises caps, and resumes paused scopes. The new-budget form defaults to warn; choose hard cap to opt into auto-pause.

​

Steps

​

1. Read the existing budgets

The Budgets section lists every budget row, newest-updated first. The header pill shows the count (N budgets). Each row shows:

• The scope badge (agent · mission · team · tenant) and its scope id.
• The mode pill (warn or cap) and the lifecycle status pill (active → success/mint, soft_capped → warning/amber, paused → error/red).
• A spent / cap line in dollars plus a percentage and a progress bar.
• If a cap-mode budget is active but already at or over its limit, a will re-pause pill: the next recorded spend re-pauses it, so raise the cap to make progress.

If the budgets load fails on the first fetch, the section shows a Couldn't load budgets alert with a Retry link. A transient poll failure after a good load keeps the last good snapshot.

​

2. Create a budget

Use the inline create form at the bottom of the Budgets section:

1. Pick the scope (agent / mission / team / tenant).
2. Type the scope id, the id of the agent, the root (mission) task, the team, or the tenant the budget applies to.
3. Enter the limit in dollars (the placeholder reads warn at $ or cap $ depending on the mode).
4. Pick the mode (warn only or hard cap).
5. Click Set budget.

The form converts dollars to cents (Math.round(v * 100)) and POSTs { scope, scopeId, limitUsdCents, mode }. A new scope starts at spent 0 / active.

​

3. Raise a cap

Every budget row has a new cap $ input and a Set cap button. Enter a new dollar amount and click Set cap; it POSTs the same /api/governance/budgets route with the new limitUsdCents, preserving the row’s mode. Re-setting the limit recomputes status from the existing spend, so raising the cap above the current spend un-pauses a paused scope.

​

4. Resume a paused scope

A cap-mode budget that crossed 100% reads paused, and a Resume button appears on its row. Click it to force the scope back to active (a human override). The resume route accepts an optional graceUsdCents body to raise the cap above current spend in one call; the panel’s Resume button sends a bare resume (no grace).

​

5. Review the enforced-in-code caps

The Caps (enforced in code) section is informational; these limits are enforced in the orchestrator/executor, not editable from the UI: A cap hit is logged to the audit log below (event type cap_hit).

​

6. Resolve approvals

The Approval queue section embeds the shared ToolApprovalQueue, the same queue and resolve UX as the Approvals panel, so there is one resolve path. A pending tool-call or delegation approval shows the tool name, an expiry countdown, the reason, and an args summary, with three actions:

• Allow Once: approve this one call.
• Always: allow-always for this scope.
• Deny: reject it.

These POST to /api/tools/approvals/:id/resolve. The queue polls /api/tools/approvals?status=pending every 3 seconds; when empty it shows a “No pending approvals” empty state.

​

7. Search the audit log

The Audit log section reads the append-only forensic log: installs, approvals, tool calls, budget events, cap hits, verifications, and circuit breaks. There is no write endpoint; subsystems write the audit in-process as agents run. Filter it three ways (all combine):

• Agent id: a free-text input matching agentId.
• Event type: a dropdown over install, approval, tool_call, budget, cap_hit, verification, circuit_break (or all events).
• Since window: All / 1h / 24h / 7d pills (sent as since = Date.now() - window).

Each row shows the local time, the event-type badge, a truncated agent id, and the redacted summary. The panel requests limit: 200; the route caps limit at 1000. Credential-shaped keys in the summary are masked at the rendering boundary (redact-on-display) on top of the write-time scrub.

​

Options / variations

The budget routes accept fields the panel does not surface directly:

​

Verify it worked

• A new/raised budget appears (or updates) in the Budgets list within the 5-second poll; its spent / cap line, percentage, and mode/status pills reflect the change. You can confirm directly with curl http://localhost:18790/api/governance/budgets (use your resolved API port).
• A resumed scope flips its status pill from paused (red) back to active (mint). If it shows will re-pause, raise the cap.
• A resolved approval disappears from the Approval queue on the next 3-second refetch.
• An audit entry for the action lands in the Audit log, e.g. setting/raising a cap-mode budget that later crosses produces budget events; a cap hit produces cap_hit; a verification produces verification.

​

Troubleshooting

​

Related

• Governance, budgets, the kill-switch, circuit breakers, caps, and approvals explained
• /api/governance reference, full request/response shapes for budgets, resume, audit, and delegation approval
• Approvals, the same tool-approval queue surfaced standalone
• Cost and budgets, the cost dashboard alongside budgets
• Production defaults, why track-and-warn is the default posture