Bug type
Regression (worked before, now fails)
Beta release blocker
No
Summary
Isolated agentTurn cron jobs intermittently fail to spawn a session at all. The cron scheduler fires, waits 6–17 minutes, then reports "cron: job execution timed out" with no sessionId, no model, and no token usage — indicating the isolated worker never launched. The failure is correlated with main session inactivity: jobs succeed reliably when the main session is active, and fail intermittently after it goes idle.
Steps to reproduce
- Gateway is freshly started or main session is actively receiving messages → jobs complete successfully in 15–55 seconds ✅
- Main session goes idle for ~1 hour → next 2–5 isolated cron runs fail with Type B timeout ❌
- Main session receives a message → isolated cron run succeeds again ✅
This cycle repeats consistently. Post-update (2026.4.9), the job ran successfully 5 consecutive times while the main session was active, then failed twice once the session went idle.
Expected behavior
Isolated agentTurn cron jobs must not fail to spawn a session.
Actual behavior
Isolated agentTurn cron jobs intermittently fail to spawn a session at all
OpenClaw version
2026.4.9
Operating system
macOS 26.4 (arm64)
Install method
nom
Model
anthropic/claude-sonnet-4.6
Provider / routing chain
local loopback
Additional provider/model setup details
No response
Logs, screenshots, and evidence
**Affected job**
- Job ID: `035c3507-2f85-4d43-9a68-20ec3a3599f1`
- `sessionTarget: "isolated"`, `payload.kind: "agentTurn"`, `lightContext: true`
- Schedule: every 55 minutes
Two distinct failure modes appear in the run history:
**Type A — Anthropic API timeout** (has `sessionId`, model info present):
error: "Request timed out before a response was generated"
durationMs: ~63000–176000
The session launched but Anthropic timed out. This was partially addressed by adding `lightContext: true`.
**Type B — Session never spawns** (no `sessionId`, no model):
error: "cron: job execution timed out"
durationMs: 267000–1077000 (~4–18 minutes)
The gateway never launched the isolated session. No Anthropic call was made. This is the dominant failure mode and is **not fixed** by `lightContext: true` or updating to 2026.4.9.
**Sample run history (chronological, today)**
| Time (EDT) | Status | Duration | sessionId? |
|---|---|---|---|
| 07:54 | ✅ ok | 14s | yes |
| 08:23 | ❌ timeout | 794s | no |
| 09:18 | ❌ timeout | 963s | no |
| 10:11 | ❌ timeout | 267s | no |
| 10:57 | ❌ timeout | 1018s | no |
| 12:29 | ✅ ok | 15s | yes |
| 14:11 | ✅ ok | 54s | yes |
| 15:06 | ✅ ok | 21s | yes |
| 16:01 | ✅ ok | 32s | yes |
| 17:00 | ✅ ok | 27s | yes |
| 17:57 | ❌ timeout | 991s | no |
| 18:52 | ❌ timeout | 404s | no |
Active session periods align exactly with successful runs. Idle periods align with Type B failures.
Impact and severity
No response
Additional information
No response
Bug type
Regression (worked before, now fails)
Beta release blocker
No
Summary
Isolated
agentTurncron jobs intermittently fail to spawn a session at all. The cron scheduler fires, waits 6–17 minutes, then reports"cron: job execution timed out"with nosessionId, nomodel, and no token usage — indicating the isolated worker never launched. The failure is correlated with main session inactivity: jobs succeed reliably when the main session is active, and fail intermittently after it goes idle.Steps to reproduce
This cycle repeats consistently. Post-update (2026.4.9), the job ran successfully 5 consecutive times while the main session was active, then failed twice once the session went idle.
Expected behavior
Isolated
agentTurncron jobs must not fail to spawn a session.Actual behavior
Isolated
agentTurncron jobs intermittently fail to spawn a session at allOpenClaw version
2026.4.9
Operating system
macOS 26.4 (arm64)
Install method
nom
Model
anthropic/claude-sonnet-4.6
Provider / routing chain
local loopback
Additional provider/model setup details
No response
Logs, screenshots, and evidence
Impact and severity
No response
Additional information
No response