Now: If the seat-history request succeeds but the standing request fails, fetchDetail returns standing: null as a successful bundle. AgentDetail replaces previously loaded work, removes its progress and step list, says “Updated”, and offers no failure notice or retry.

Change to: Treat a rejected or invalid standing response as a failed detail refresh instead of returning standing: null. Throw through the existing AgentDetail error handler so it retains the previous bundle, shows its saved timestamp and “Could not refresh this agent”, and exposes the existing Try again action.

Why: A tracker must distinguish unavailable current-work data from a successful update and preserve known tasks during a transient failure. This concerns the repository’s handling of a partial request failure, not the public service’s availability.

Now: The tool opens with “The swarm at work.” and “Find an agent. Follow what happens next.”, then labels the office “A little office. Real work.” and adds “Every agent has a story. Pick one.” beside the directory.

Change to: Use “Agent overview” for the page heading and “Featured agents” for the office heading; delete the hero subtitle and directory slogan. Keep the search label and office selection instruction, and let the search, metrics, and office move up into the deleted subtitle’s space.

Why: This is a dashboard for finding agents and checking work; the review’s copy bar calls for plain task headings and no voice lines in tools. The artwork already supplies the brief’s playful character.