Bug type
Regression (worked before, now fails)
Beta release blocker
No
Summary
cron jobs using isolated sessions (sessionTarget="isolated") intermittently fail at startup with "model-call-started" timeout. Switching model/provider does not help. Switching to main session (sessionTarget="main", systemEvent) avoids the issue entirely.
Steps to reproduce
- Configure a cron job with sessionTarget="isolated" and payload.kind="agentTurn"
- Observe that the job intermittently fails with error: "cron: job execution timed out (last phase: model-call-started)"
- The timeout happens before any model call starts (no API calls in logs)
- Switching the model to a completely different provider (e.g. DeepSeek → Kimi) does not fix it
- Converting the same job to sessionTarget="main" + payload.kind="systemEvent" works reliably every time
Expected behavior
Isolated cron sessions should initialize reliably. If the model cannot be reached, the error should reflect the actual API failure (e.g. HTTP timeout, auth error) rather than a phantom "model-call-started" phase timeout before any real network call.
Actual behavior
cron jobs with sessionTarget="isolated" intermittently fail (~4/4 attempts for one weekly job, occasionally for others) with:
cron: job execution timed out (last phase: model-call-started)
The error consistently happens at the "model-call-started" phase, before the model provider is even contacted.
Evidence:
- Multiple jobs affected: "挖新技能-每周" (new skill search), "每日AI新闻简报" (daily news digest)
- 4 consecutive failures on one job before migration
- Same model/provider (deepseek/deepseek-v4-flash with moonshot/kimi-k2.6 fallback) works fine in main session
- 8 test runs with the same model in main session completed in <12s each
- systemEvent (main session) mode works every time
OpenClaw version
2026.6.6 (8c802aa)
Operating system
macOS 26.5.1 (arm64)
Install method
npm global
Model
deepseek/deepseek-v4-flash (Primary), moonshot/kimi-k2.6 (Fallback)
Provider / routing chain
openclaw -> deepseek (primary) / moonshot (fallback). No proxy or gateway in between.
Additional provider/model setup details
No response
Logs, screenshots, and evidence
Impact and severity
Affected: Any user with cron jobs configured as agentTurn (sessionTarget="isolated")
Severity: Medium - jobs silently fail with no output
Frequency: Intermittent (observed ~4/4 on one weekly job, occasionally on others)
Consequence: Scheduled automations (skill discovery, news digests, etc.) fail without user visibility. The only workaround is converting all isolated jobs to main session mode.
Additional information
No response
Bug type
Regression (worked before, now fails)
Beta release blocker
No
Summary
cron jobs using isolated sessions (sessionTarget="isolated") intermittently fail at startup with "model-call-started" timeout. Switching model/provider does not help. Switching to main session (sessionTarget="main", systemEvent) avoids the issue entirely.
Steps to reproduce
Expected behavior
Isolated cron sessions should initialize reliably. If the model cannot be reached, the error should reflect the actual API failure (e.g. HTTP timeout, auth error) rather than a phantom "model-call-started" phase timeout before any real network call.
Actual behavior
cron jobs with sessionTarget="isolated" intermittently fail (~4/4 attempts for one weekly job, occasionally for others) with:
The error consistently happens at the "model-call-started" phase, before the model provider is even contacted.
Evidence:
OpenClaw version
2026.6.6 (8c802aa)
Operating system
macOS 26.5.1 (arm64)
Install method
npm global
Model
deepseek/deepseek-v4-flash (Primary), moonshot/kimi-k2.6 (Fallback)
Provider / routing chain
openclaw -> deepseek (primary) / moonshot (fallback). No proxy or gateway in between.
Additional provider/model setup details
No response
Logs, screenshots, and evidence
Impact and severity
Affected: Any user with cron jobs configured as agentTurn (sessionTarget="isolated")
Severity: Medium - jobs silently fail with no output
Frequency: Intermittent (observed ~4/4 on one weekly job, occasionally on others)
Consequence: Scheduled automations (skill discovery, news digests, etc.) fail without user visibility. The only workaround is converting all isolated jobs to main session mode.
Additional information
No response