[FEATURE] Add liveness/progress signal for background subagents (Task tool)

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

Summary

When launching subagents in the background (Task tool / /tasks), there is no intermediate signal between "launched" and "completed/failed". A subagent stuck retrying a failing tool call, looping, or silently stalled looks identical — from the user's point of view — to one that is simply thinking normally. The only feedback today is a final notification (success or error); nothing in between.

Current behavior

• Each subagent gets a visible name/ID and shows up in /tasks.
• On completion (success or failure), a notification is sent.
• While running: no heartbeat, no progress indicator, no way to distinguish "working normally" from "stuck in a loop / repeatedly failing a tool call / hung".

Problem this causes

With multiple subagents running in parallel (common in multi-agent orchestration setups), a user has no way to tell — without interrupting or waiting for the final notification — whether a given agent is:

1. Making normal progress on a long task, or
2. Stuck retrying the same failing action, or
3. Effectively hung.

This makes it hard to decide when to let something keep running vs. intervene.

Proposed Solution

Suggested improvement

Surface a lightweight liveness/progress indicator per running subagent, for example:

• A last-activity timestamp or "last tool call" summary visible in /tasks.
• A simple heartbeat state (e.g., "active" vs "idle/stalled for Xs") derived from tool-call cadence.
• Optionally, a warning signal when an agent repeats the same failing tool call N times in a row.
• a button to interrupt or add new consign for the agent (like the orchestrator do, but manually)

Doesn't need to be verbose streaming output — just enough signal to distinguish "still working" from "likely stuck."

Alternative Solutions

_No response_

Priority

Medium - Would be very helpful

Feature Category

Developer tools/SDK

Use Case Example

_No response_

Additional Context

Context

Reported from a multi-agent Claude Code setup orchestrating several specialized subagents in parallel on a single project.