Feature Request: Reliable Subagent Spawn with Built-in Polling

Use Case

When spawning subagents for long-running tasks, agents need:

• Automatic polling for completion (not manual checks)
• Timeout handling
• Auto-retry on stall/crash
• Clear failure escalation

Currently, agents must manually poll sessions_history or risk losing track of subagent results.

Proposed API

Option A: Extend sessions_spawn with wait options:

{
  "tool": "sessions_spawn",
  "arguments": {
    "task": "...",
    "label": "...",
    "wait": true,              // NEW: block until complete
    "timeout": 300,            // NEW: total timeout (seconds)
    "checkInterval": 10,       // NEW: poll frequency (seconds)
    "maxRetries": 2,           // NEW: respawn on stall
    "onTimeout": "escalate"    // NEW: "escalate" | "return_partial"
  }
}

Option B: New tool sessions_spawn_and_wait with same arguments (mandatory wait).

Implementation Notes

• Gateway already tracks subagent sessions
• Internal polling loop won't expose complexity to agents
• Backward compatible (wait=false or new tool)
• Returns final result or structured error with logs

Benefits

• Reliability becomes default, not optional
• Eliminates "did my subagent finish?" uncertainty
• Handles respawn logic consistently
• Matches "DO BEFORE ASK" agent philosophy

Related

Pattern documented in agent workspaces using helper scripts, but native support would be cleaner.