Bug type

Regression (worked before, now fails)

Summary

sessions_yield (introduced in v2026.3.12 via PR #36537) immediately terminates isolated cron sessions instead of pausing and waiting for a follow-up event (like a sub-agent completion). The description says it should "end the current turn and carry a follow-up payload into the next session turn", but in isolated cron sessions there is no next turn, so the session just dies.

Steps to reproduce

1. Create a cron job with sessionTarget: "isolated" and payload.kind: "agentTurn"
2. In the agent turn, spawn a sub-agent via sessions_spawn (mode: "run")
3. Call sessions_yield to wait for the sub-agent completion event
4. The isolated session terminates immediately (~30 seconds total runtime)
5. The sub-agent runs and completes, but the orchestrator session is already dead
6. Steps after the sub-agent (drafting, posting, notifications) never execute

Expected behavior

sessions_yield should keep the isolated cron session alive and waiting for the sub-agent completion event to arrive, then resume execution. Or at minimum, it should be documented that sessions_yield is incompatible with isolated sessions.

Actual behavior

The session terminates immediately after sessions_yield is called. The cron scheduler reports lastStatus: "ok" with very short durations (28-36 seconds) even though the full pipeline should take 3-5 minutes.

OpenClaw version

2026.3.13 (also reproduced on 2026.3.11 which bundles 3.12 changes)

Operating system

Linux (Debian 13)

Install method

No response

Model

anthropic/claude-sonnet-4-6 (subagent), anthropic/claude-opus-4-6 (orchestrator)

Provider / routing chain

openclaw -> anthropic direct

Config file / key location

No response

Additional provider/model setup details

No response

Logs, screenshots, and evidence

Impact and severity

Affected: Any isolated cron session using sessions_yield with sub-agent orchestration
Severity: High - blocks multi-step pipeline workflows entirely
Frequency: 100% repro - always terminates immediately
Consequence: Multi-step orchestrator pipelines (spawn -> draft -> post -> notify) fail silently. Cron reports lastStatus: "ok" masking the failure. Sub-agents run but their results are never processed.

Additional information

Workaround:
After spawning a sub-agent, simply end the response without calling any tool. The sub-agent completion event arrives as the next message naturally. Do not call sessions_yield in isolated cron sessions.

Additional context:
This affects any orchestrator pattern where an isolated cron session spawns sub-agents and needs to continue processing after they complete. The multi-step pipeline pattern (spawn discovery -> draft -> post -> notify) breaks entirely because the orchestrator dies at the first yield.

Previously working on v2026.2.17 (before sessions_yield existed, agent used natural turn waiting).