[P1] Build a Fleet Operations Console for 20–100+ Agents

[0YHR0]

Problem

The current Console is strong as a task/demo visualization, while the Office is engaging as an
experience. Neither makes the primary control-plane value obvious to an operator managing 20–100+
heterogeneous Agents.

Proposal

Create a clean Fleet Operations Console as the default administrator experience. Office/Mission Map
remain optional visualizations. The main view must answer:

• what Agents, versions, runtimes, owners, and capabilities exist;
• what is running, queued, degraded, blocked, or revoked;
• which actions/tools were requested, permitted, denied, approved, or outcome-unknown;
• current spend/quota and failure/recovery signals;
• where operator attention is required now.

Acceptance criteria

• One fleet inventory supports search/filter by status, runtime, capability, owner, risk, and
  version.
• Agent detail joins deployments, active Runs, permissions, credential/Permit posture, cost,
  failures, and recent governed actions.
• Attention queue unifies approvals, unknown outcomes, repeated failures, drift, and unhealthy
  runtimes with direct actions.
• Run detail presents an evidence timeline from assignment through recovery and output—not raw
  internal tables.
• Revocation/cancel/pause actions show impact and require confirmation where destructive.
• UI remains usable with seeded datasets of at least 100 Agents and 1,000 recent Runs.
• English and Chinese, keyboard navigation, reduced motion, responsive layout, and empty/loading/
  error states are covered by tests.
• The Office links to the same resources but is not required to administer the system.

Non-goals

• Rebuilding the product as a game.
• Adding a generic no-code workflow builder in this issue.
• Hiding governance details behind only aggregate scores.