Problem

When an orchestrator agent spawns multiple sub-agents (e.g., a swarm of 8-15 coding agents), the orchestrator session goes idle immediately after dispatching. No progress updates are posted to the delivery channel until a human manually pings the session.

This is a fundamental architectural issue, not a behavioral one. The agent cannot post updates because it is not running — its session is dormant.

Reproduction

1. Configure an agent with instructions to post progress updates every 60 seconds
2. Have it spawn 10+ sub-agents via sessions_spawn(mode: run)
3. Observe the delivery channel (Slack, Telegram, webchat)
4. Expected: Periodic updates as sub-agents complete (5/15 done, 10 running)
5. Actual: Complete silence until the human sends a message, at which point the orchestrator wakes, processes all queued completions at once, and posts a batched update

Root Cause Analysis

After sessions_spawn, the orchestrator session has no pending work. OpenClaw session lifecycle puts it to sleep. Sub-agent completion events queue up but the orchestrator is not awake to process them.

The session only wakes on:

• Inbound user message (human ping)
• Cron system event
• Another session sending it a message via sessions_send

None of these happen automatically when a sub-agent completes.

Impact

• User experience: Users must repeatedly ping to get status updates on multi-agent work
• Situational awareness: Failed sub-agents go unnoticed for minutes/hours
• Trust erosion: Users lose confidence when the system goes silent during active work

Proposed Solutions

Option A: Completion-triggered session wake (preferred) - When a sub-agent completes, auto-wake the parent session.

Option B: Push-based completion hooks - Allow registering callbacks when spawning.

Option C: Built-in swarm progress tracking - First-class swarm concept with auto-reporting.

Environment

• OpenClaw v2026.3.13 / macOS arm64 / 45 agents / ACP runtime with Codex