Sub-Agents

Monitor spawned sub-agent tasks, outputs, costs, and lineage

The Sub-Agents view shows all recently spawned sub-agent runs. Sub-agents are created when an agent delegates tasks to child agents during execution — for example, a research agent spawning a summarizer. This view lets you monitor their status, review results, and terminate stuck runs.

Route

• /subagents — opens the sub-agents monitoring view

Backing RPC and SSE

• subagent.list — runs from the last 60 minutes (with status, depth, parent, cost, tokens, steps)
• subagent.kill(spawnId) — cancel a running or queued sub-agent
• subagent.steer(spawnId, message) — send a steering message to a running sub-agent
• SSE events subagent:spawned and subagent:status drive a debounced auto-reload so you see status changes without refreshing

What You See

Run List

The main area displays sub-agent runs from the last 60 minutes as a vertical list of cards. Each run card shows:

Element	Description
Status	Colored tag showing the current state: running (blue), queued (gray), completed (green), failed (red), killed (yellow)
Agent ID	The identifier of the sub-agent that was spawned
Task	The task description given to the sub-agent (truncated to 80 characters, full text on hover)
Started	Relative timestamp of when the run began
Duration	How long the run took (shown for completed runs)
Depth	The spawn depth level (1 = direct child, 2 = grandchild, etc.)
Tokens	Total tokens used (shown for completed runs with results)
Cost	Total cost in USD (shown for completed runs with results)
Steps	Number of execution steps completed (shown for completed runs)

Failed runs display an error message in a red-highlighted section below the metadata.

Kill Action

Running or queued sub-agents display a Kill button. Clicking it opens a confirmation dialog warning that the parent session will receive an error result. This is useful for terminating sub-agents that are stuck or consuming excessive resources.

Real-Time Updates

The view subscribes to SSE lifecycle events. When a sub-agent starts, completes, or fails, the list automatically refreshes via a debounced reload. This means you see status changes without manually refreshing the page.

Empty State

If no sub-agent runs occurred in the last 60 minutes, the view shows an empty state message. Sub-agent runs are only retained in the runtime list for recent activity monitoring — historical data is available through session logs.

Common Tasks

1. Monitor active sub-agents

   Open the Sub-Agents view. Running sub-agents appear with a blue status tag. Watch for status changes as they complete or fail.

2. Check sub-agent cost

   Look at the Tokens and Cost fields on completed run cards. These show the total resource consumption for each sub-agent task.

3. Kill a stuck sub-agent

   Find the running sub-agent in the list, click the Kill button, and confirm in the dialog. The parent agent’s session will receive an error result for that task.

4. Review sub-agent errors

   Failed runs display their error message in a red section below the metadata. Check the error text and the parent agent’s session for context.

Related Pages

Sub-Agent Lifecycle

How sub-agents are spawned, managed, and cleaned up.

Execution Graphs

Multi-agent pipelines that orchestrate sub-agent tasks.

Sessions View

View conversation histories including sub-agent sessions.