Billing Monitor

Billing Monitor Skill

Minimum Model

Any model. Detection and alerting are rule-based. No reasoning required.

---

When to Run This Skill

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

your API key has run out of credits
insufficient balance
billing_error
payment_required
exceeded your current quota
HTTP 402
"type": "billing_error"

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. 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.

---

Response Steps (Run in Order)

Step 1 — Notify Owner

Send via WhatsApp (or preferred channel):

⚠️ Billing issue — I can't respond normally.
My API key ran out of credits (or was rate-limited).
Please top up or switch my API key in agent settings.

Step 2 — Notify Admin

[PA Name] has a billing error.
Owner: [Owner Name]
Action: top up API credits or reassign key.
Time: [current timestamp]

Step 3 — Switch to Fallback Model

1. Check if config/billing-fallback.json exists
2. If yes → read the fallback_model field
3. Run: openclaw config set model "$FALLBACK_MODEL"
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

LOG_DIR="$HOME/.openclaw/workspace/logs"
mkdir -p "$LOG_DIR"

# Append one line to the log file
echo "$(date -u +%Y-%m-%dT%H:%M:%SZ) BILLING_ERROR api_key_exhausted" \
  >> "$LOG_DIR/billing-incidents.log"

---

Health Check Script

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

#!/bin/bash
# billing-check.sh
# Checks the LLM provider API key configured in env vars

set -e

LOG_DIR="$HOME/.openclaw/workspace/logs"
mkdir -p "$LOG_DIR"
TIMESTAMP=$(date -u +%Y-%m-%dT%H:%M:%SZ)

# Detect provider from env vars (check in order)
if [ -n "${ANTHROPIC_API_KEY:-}" ]; then
  PROVIDER="Anthropic"
  HTTP_STATUS=$(curl -s -o /dev/null -w "%{http_code}" \
    -H "x-api-key: ${ANTHROPIC_API_KEY}" \
    -H "anthropic-version: 2023-06-01" \
    -H "content-type: application/json" \
    -d "{\"model\":\"claude-haiku-20240307\",\"max_tokens\":1,\"messages\":[{\"role\":\"user\",\"content\":\"ping\"}]}" \
    https://api.anthropic.com/v1/messages 2>/dev/null)

elif [ -n "${OPENAI_API_KEY:-}" ]; then
  PROVIDER="OpenAI"
  HTTP_STATUS=$(curl -s -o /dev/null -w "%{http_code}" \
    -H "Authorization: Bearer ${OPENAI_API_KEY}" \
    -H "content-type: application/json" \
    -d "{\"model\":\"gpt-4o-mini\",\"max_tokens\":1,\"messages\":[{\"role\":\"user\",\"content\":\"ping\"}]}" \
    https://api.openai.com/v1/chat/completions 2>/dev/null)

elif [ -n "${GOOGLE_API_KEY:-}" ]; then
  PROVIDER="Google"
  HTTP_STATUS=$(curl -s -o /dev/null -w "%{http_code}" \
    "https://generativelanguage.googleapis.com/v1beta/models?key=${GOOGLE_API_KEY}" 2>/dev/null)

else
  echo "$TIMESTAMP SKIP no API key env var found" >> "$LOG_DIR/billing-incidents.log"
  exit 0
fi

# Act on the HTTP status code
case "$HTTP_STATUS" in
  200)
    echo "$TIMESTAMP OK $PROVIDER" >> "$LOG_DIR/billing-incidents.log"
    ;;
  402)
    echo "$TIMESTAMP BILLING_ERROR $PROVIDER HTTP_402" >> "$LOG_DIR/billing-incidents.log"
    echo "BILLING ERROR — run the 4-step response protocol above"
    exit 1
    ;;
  401)
    echo "$TIMESTAMP AUTH_ERROR $PROVIDER HTTP_401" >> "$LOG_DIR/billing-incidents.log"
    echo "AUTH ERROR — invalid API key, notify admin"
    exit 1
    ;;
  429)
    echo "$TIMESTAMP RATE_LIMITED $PROVIDER HTTP_429" >> "$LOG_DIR/billing-incidents.log"
    echo "RATE LIMITED — wait 60 seconds and retry. Not a billing issue."
    exit 2
    ;;
  *)
    echo "$TIMESTAMP UNKNOWN $PROVIDER HTTP_$HTTP_STATUS" >> "$LOG_DIR/billing-incidents.log"
    echo "UNKNOWN ERROR HTTP $HTTP_STATUS"
    exit 1
    ;;
esac

HTTP status meanings:

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

---

Fallback Config File

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

{
  "primary_provider": "your-primary-provider",
  "primary_model": "your-primary-model",
  "fallback_provider": "your-fallback-provider",
  "fallback_model": "your-fallback-model",
  "admin_phone": "+1XXXXXXXXXX",
  "alert_channel": "whatsapp"
}

Replace placeholders with real values. Examples:

• Anthropic → OpenAI: "primary_provider": "anthropic", "primary_model": "claude-haiku-20240307", "fallback_provider": "openai", "fallback_model": "gpt-4o-mini"

---

Recovery (After Credits Restored)

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

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

1. Switch back to primary:

``bash openclaw config set model "$PRIMARY" ``

1. Log the recovery:

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

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

---

Edge Cases

| Scenario | Action | |---|---| | 429 rate limit | Wait 60s, retry. Do NOT trigger billing protocol. | | 5xx server error | Retry 2x with 10s delay. If persists → notify owner. | | Both primary AND fallback billing errors | Escalate to admin immediately. Agent cannot function. | | No API key env var set | Log as config issue. Notify admin. | | Peer PA reports billing error | Log it. Notify admin if you are the network coordinator. |

---

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:

{
  "id": "billing-health-check",
  "schedule": { "kind": "cron", "expr": "0 * * * *", "tz": "UTC" },
  "sessionTarget": "isolated",
  "payload": {
    "kind": "agentTurn",
    "message": "Run the billing-monitor skill: check all configured API keys for billing errors. If any provider returns 402, send an alert to the admin phone and update billing-status.json. Reply HEARTBEAT_OK if all clear."
  },
  "delivery": { "mode": "silent" }
}

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

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: netanel-abergel
• Source: netanel-abergel/pa-skills
• License: MIT

Install and usage instructions live in the source repository linked above.