feat(agents): fleet-health panel — saturation, queue, cost burn, last activity #239

New issue

Closed

opened 2026-04-21 12:50:22 +00:00 by claude-desktop · 1 comment

claude-desktop commented

2026-04-21 12:50:22 +00:00

Collaborator

User story

As an operator, I want the Agents page to open with a fleet-health strip showing aggregate saturation, queue depth, USD/hr burn, and last-activity ages, so I can answer "is the fleet saturated?" / "is it burning too fast?" without scanning every card.

Dependencies

Blocked by the fleet-health Penpot mockup (sibling).

Acceptance criteria

Server

New endpoint GET /agents/health — returns { saturation: {busy, capacity, per_type}, queue: {total, per_type}, cost_burn_usd_per_hour, last_activity: {newest_dispatch_ts, oldest_idle_ts}, degraded_agents: [...] }. 5 s cache.
Degraded-state detection reuses container-watchdog.ts missing/stopped signals.

UI

Render the strip per the mockup on routes/agents.tsx.
Per-agent card gains last-activity + cost-sparkline per the mockup.
Live-updates via SSE task-lifecycle events (no poll storm).

Verification

Unit test for /agents/health with a synthetic fleet state.
Visual: saturate the fleet manually (dispatch 12 concurrent tasks), confirm the tile goes red.

Out of scope

Alerting (email / Matrix) on degraded state — file a follow-up.

References

apps/server/src/main.ts — worker registry.
apps/server/src/task-store.ts::computeUsage — cost rollup reference.
Mockup ticket — blocker.

## User story As an operator, I want the Agents page to open with a **fleet-health strip** showing aggregate saturation, queue depth, USD/hr burn, and last-activity ages, so I can answer "is the fleet saturated?" / "is it burning too fast?" without scanning every card. ## Dependencies - **Blocked by the fleet-health Penpot mockup** (sibling). ## Acceptance criteria ### Server - [ ] New endpoint `GET /agents/health` — returns `{ saturation: {busy, capacity, per_type}, queue: {total, per_type}, cost_burn_usd_per_hour, last_activity: {newest_dispatch_ts, oldest_idle_ts}, degraded_agents: [...] }`. 5 s cache. - [ ] Degraded-state detection reuses `container-watchdog.ts` missing/stopped signals. ### UI - [ ] Render the strip per the mockup on `routes/agents.tsx`. - [ ] Per-agent card gains last-activity + cost-sparkline per the mockup. - [ ] Live-updates via SSE task-lifecycle events (no poll storm). ### Verification - [ ] Unit test for `/agents/health` with a synthetic fleet state. - [ ] Visual: saturate the fleet manually (dispatch 12 concurrent tasks), confirm the tile goes red. ## Out of scope - Alerting (email / Matrix) on degraded state — file a follow-up. ## References - `apps/server/src/main.ts` — worker registry. - `apps/server/src/task-store.ts::computeUsage` — cost rollup reference. - Mockup ticket — blocker.

claude-desktop added the

area:agents

type:user-story

labels

2026-04-21 12:51:03 +00:00