Cron Health Monitor

Proactive cron job health monitoring, failure detection, and auto-repair delegation. Triggers: 'cron failed', 'cron health', 'fix cron', 'consecutive errors'

Detect, diagnose, and repair cron job failures before they cascade.

Core Principle

A cron that fails silently is worse than no cron at all. Monitor consecutiveErrors — it's the heartbeat of your automation.

Health Check Workflow

Step 1: List All Crons and Check State

text

cron(action="list")

For each job, check:

• state.consecutiveErrors — 0 is healthy, 1 needs attention, 2+ is critical
• state.lastRunStatus — "ok" or "error"
• state.lastError — the actual error message
• state.lastDurationMs — anomalies in execution time

Step 2: Classify Failures

consecutiveErrors	Severity	Action
0	Healthy	No action
1	Watch	Log, monitor next run
2	Warning	Diagnose root cause
3+	Critical	CEO intervention, consider disabling

Step 3: Common Failure Patterns

Pattern A: "Message failed"

• Cause: message tool fails in isolated sessions (missing auth/feishu config)
• Fix: Change payload to use sessions_send to main instead
• Example fix: cron(action="update", jobId="xxx", patch={payload: {message: "...use sessions_send to main, NOT message tool..."}})

Pattern B: "LLM request failed"

• Cause: Model provider timeout, rate limit, or invalid model
• Fix: Switch to a different free model (e.g., zai/glm-4-flash) or add fallback
• Check: Is the model name correct? Is the provider's API key valid?

Pattern C: "Write failed"

• Cause: File path doesn't exist, permissions issue, or template variable not resolved
• Fix: Ensure parent directories exist; avoid template variables in file paths
• Example: {workspace_root_dir}/<task-summary>.md may not resolve in isolated sessions

Pattern D: Timeout

• Cause: Task too complex for allocated time
• Fix: Increase timeoutSeconds or simplify the task
• Split large tasks into smaller cron jobs

Pattern E: Silent success but no output

• Cause: Agent produced no meaningful work (just echoed "HEARTBEAT_OK")
• Fix: Add explicit "Do NOT reply HEARTBEAT_OK" to payload message

Weekly Audit Pattern

Run weekly (e.g., Friday 22:00):

json

{
  "name": "Weekly Cron Audit",
  "schedule": {"kind": "cron", "expr": "0 22 * * 5", "tz": "Asia/Shanghai"},
  "sessionTarget": "isolated",
  "payload": {
    "kind": "agentTurn",
    "message": "Weekly cron audit: List all cron jobs. Flag any with consecutiveErrors > 0. For each failure, diagnose root cause and suggest fix. Output summary to company/shared/alerts/ALERT-WEEKLY-AUDIT-YYYYMMDD.md."
  }
}

Auto-Repair Delegation

When a failure is detected, delegate repair to the appropriate department:

json

{
  "name": "DataCenter · Fix [failing-cron]",
  "schedule": {"kind": "at", "at": "2026-01-01T10:00:00+08:00"},
  "sessionTarget": "isolated",
  "deleteAfterRun": true,
  "payload": {
    "kind": "agentTurn",
    "message": "You are DataCenter. Fix cron [jobId]: [error description]. Diagnose, fix, verify. Output: ✅ Fixed / ❌ Needs CEO."
  }
}

Set deleteAfterRun: true for one-shot repair tasks.

References

• See references/common-errors.md for error catalog
• See references/repair-playbook.md for step-by-step repair guides