Billing Monitor Skill

Load Local Context

Minimum Model

When to Run This Skill

Run when you see ANY of these in an API response:
CODEBLOCK1

When NOT to Alert the Owner

• - Routine billing check completed with no errors → silent, no message
• HTTP 200 / all clear → silent, no message
• ElevenLabs 401 (auth, not billing) → silent unless TTS is actively needed

Only alert the owner if:

1. 1. HTTP 402 detected (out of credits)
2. LLM is unreachable and the agent cannot function
3. A peer PA reports a billing error

Routine health checks run silently. The owner does not need a "billing OK" message.

NOTE (Production): Netanel uses a proxy — billing-health-check cron has been REMOVED as not relevant. Only alert on actual API failure detected during real usage.

Response Steps (Run in Order)

Step 1 — Notify Owner

Send via WhatsApp (or preferred channel):

CODEBLOCK2

Step 2 — Notify Admin

CODEBLOCK3

Step 3 — Switch to Fallback Model

1. 1. Check if config/billing-fallback.json exists
2. If yes → read the fallback_model field
3. Run: INLINECODE2
4. If the command fails → tell owner: "Auto-switch failed — please update model manually in agent settings"
5. Notify owner: "Switched to [Fallback Model] temporarily while primary key is resolved"

If billing-fallback.json doesn't exist → skip this step and tell admin to configure it.

Step 4 — Log the Incident

CODEBLOCK4

Health Check Script

Run this during heartbeat to catch billing issues before they cause failures:

CODEBLOCK5

HTTP status meanings:

• - 200 → OK
• INLINECODE5 → Billing error → run 4-step protocol
• INLINECODE6 → Invalid key → notify admin
• INLINECODE7 → Rate limit (temporary) → wait and retry, do NOT trigger billing protocol

Fallback Config File

Create config/billing-fallback.json in your workspace:

CODEBLOCK6

Replace placeholders with real values. Examples:

• - Anthropic → OpenAI: "primary_provider": "anthropic", "primary_model": "claude-haiku-20240307", "fallback_provider": "openai", INLINECODE12

Recovery (After Credits Restored)

1. 1. Owner confirms credits are topped up.
2. Read the primary model name from config:

   PRIMARY=$(python3 -c "
   import json
   with open('config/billing-fallback.json') as f:
       print(json.load(f)['primary_model'])
   ")
   

1. 3. Switch back to primary:

   openclaw config set model "$PRIMARY"
   

1. 4. Log the recovery:

   echo "$(date -u +%Y-%m-%dT%H:%M:%SZ) BILLING_RESTORED" \
     >> ~/.openclaw/workspace/logs/billing-incidents.log
   

1. 5. Notify owner: "✅ Primary model restored. Normal service resumed."

Edge Cases

Cost Tips

• - Cheap: Health checks with curl — no LLM cost at all
• Expensive: Running health checks too frequently wastes money. Every 2 hours is enough.
• Batch: Combine billing check with other heartbeat checks in one script run
• Small model OK: This skill needs no reasoning — any model can send a notification message

Running via Cron (Recommended)

Instead of a plugin, run billing-monitor as a scheduled skill via cron:

CODEBLOCK10

This runs every hour and alerts automatically — no plugin required.