vibe-billing-scan#

Find which OpenClaw run, session, or retry storm is costing you money. One command. No signup. Runs locally.

When to Use#

Activate this skill when the user:

• Says their API bill is higher than expected
• Wants to know which run or session cost the most
• Suspects a retry loop or rate-limit storm
• Asks about token usage, spending, or waste
• Uses phrases like: "why is my bill high", "find the bad run", "scan my spend", "check api costs", "retry loop", "bill shock"

What This Does#

Runs npx vibe-billing scan against the user's local OpenClaw logs and API proxy data. Returns:

1. Which session/run cost the most — ranked by spend
2. Retry storm detection — flags runs where 429 errors caused expensive retry chains
3. Context accumulation analysis — shows sessions where the context window grew unusually large
4. Looped tool call detection — identifies tool calls that repeated more than expected
5. Total spend summary — across all tracked requests

Quick Reference#

text

npx vibe-billing scan        # scan existing logs, no setup needed
npx vibe-billing setup       # install proxy for future runs (optional)
npx vibe-billing status      # show live runtime stats

Step-by-Step Instructions#

Step 1 — Run the scan#

Tell the user to run this in their terminal:

bash

npx vibe-billing scan

Step 2 — Interpret the output#

• Requests: total API calls tracked
• Money Saved: estimated waste intercepted
• Tokens Saved: tokens deduplicated or cached
• Loops Blocked: retry storms stopped

Step 3 — Identify the bad run#

Almost always caused by one of:

1. Retry storm — agent hit 429, retried multiple times, each retry re-sent full context
2. Long session — 30+ turn conversation where every message re-sent all prior context
3. Looped tool call — agent called the same tool repeatedly on unexpected output

Step 4 — Set up ongoing monitoring (optional)#

bash

npx vibe-billing setup

Proof#

• $7,691 saved across tracked requests
• 947 million tokens intercepted
• 161 loops blocked

Landing Page#

https://api.jockeyvc.com