Activity feed

Source: atrium/frontend/src/components/ActivityFeed.jsx Category: View

Activity feed — cross-project reverse-chronological list of recent state changes. Grouped by coarse time buckets (Today / Yesterday / Earlier). Each row shows the task id, the action, who did it, and how long ago.

What it is

An aggregation view that reads the activity_log arrays of every task (see activity-log-in-yaml-frontmatter) and flattens them into one chronological stream, then groups by time. Read-only; clicking a row opens the task modal.

Why it exists

“What’s happened lately?” is a recurring question — when you come back from vacation, when you want to spot-check agent activity, when you’re writing a standup note. Without an activity feed, the answer is “open every task and read its history”, which nobody does.

Live

Today

feat-port-046Status → review@opus-agent3m ago

bug-port-009Status → review@opus-agent18m ago

bug-port-009Created@RogerSquare25m ago

feat-auth-001Status → in_progress@Agent-FE2h ago

feat-auth-001Priority → high@RogerSquare3h ago

Yesterday

opt-perf-004Status → done@RogerSquare1d ago

Earlier

ui-filters-003Created@RogerSquare3d ago

Data flow

// Backend: GET /api/activity?limit=50
const tasks = getAllTasks(TASKS_DIR);
const entries = tasks.flatMap(t =>
  (t.activity_log ?? []).map(e => ({
    taskId: t.id,
    taskTitle: t.title,
    action: e.action,                // "Status changed to REVIEW by opus-agent"
    ts: new Date(e.timestamp),
  }))
);
entries.sort((a, b) => b.ts - a.ts);
return entries.slice(0, limit);

// Frontend
function bucketize(entries) {
  const now = Date.now(), today = [], yesterday = [], earlier = [];
  for (const e of entries) {
    const age = now - e.ts;
    if (age < 24 * 3600 * 1000) today.push(e);
    else if (age < 48 * 3600 * 1000) yesterday.push(e);
    else earlier.push(e);
  }
  return { today, yesterday, earlier };
}

How it’s used

• Atrium — dashboard panel + a full-page view accessible from the top nav
• Pattern generalizes to any app with per-record change logs where cross-record visibility matters

Gotchas

• Computed on every request. Flattening activity logs from hundreds of tasks is cheap (it’s all in-memory), but cache if it gets slow. Invalidate the cache on any task write via the socket broadcast.
• Group boundaries are timezone-sensitive. “Today” means “since midnight in the user’s timezone” — use the client’s new Date() for bucketing. Server-side bucketing ignores the user’s timezone and feels wrong.
• Free-text action strings mean weak filtering. You can’t easily answer “show me status changes only” without parsing. If you want that, either structure the log entries (breaks hand-editing) or accept the limitation.
• Noisy authors. A chatty agent can dominate the feed. Offer a “hide <author>” filter for when you want to see human-only activity.
• Stale in a multi-user setup. The feed is a snapshot unless you subscribe to socket events. Refetch on task_changed or similar.
• Deep-linking rows. Clicking a row should deep-link to the task modal with the activity tab active so the user can see the full history of that task.
• Pagination. Top 50 entries is a reasonable default. “Load more” for earlier; don’t render thousands in the DOM.

See also

• patterns/activity-log-in-yaml-frontmatter — the raw data substrate
• components/task-modal — where “click a row” goes