爪算分析器

clawculator

Cost forensics for OpenClaw. Finds billing issues in your config, sessions, and workspace. Pure deterministic logic — no AI, no network calls, no external dependencies.

Source code is fully bundled in this skill folder. Nothing is fetched at runtime. You can audit every file before running.

Files this skill reads:

• - ~/.openclaw/openclaw.json — your OpenClaw config
• INLINECODE1 — session token usage
• INLINECODE2 — workspace root file count only (no file contents read)
• INLINECODE3 — log directory (read only, if present)

Files this skill may write (only when --md is used):

• - ./clawculator-report.md — markdown report
• Custom path via INLINECODE6

No network requests are made. No shell commands are spawned.

Session keys are truncated in all output (first 8 chars + ellipsis) to avoid exposing sensitive identifiers.

When the user types clawculator, check my costs, analyze spend, or cost report, run:

CODEBLOCK0

Return the full markdown report to the user inline.

When the user types snapshot, share my grade, cost grade, clawculator snapshot, or what's my grade, run:

CODEBLOCK1

Return the terminal card output to the user. This shows their grade (A+ to D), cost range, setup complexity, and findings — designed to be screenshot and shared.

Flags

• - --snapshot — shareable grade card (terminal output, screenshot-ready)
• INLINECODE17 — write markdown report and print to stdout
• INLINECODE18 — machine-readable JSON to stdout
• INLINECODE19 — custom output path for INLINECODE20
• INLINECODE21 — full usage

What it catches

• - 💓 Heartbeat running on paid model instead of Ollama
• 🔧 Skill polling loops on paid model
• 📱 WhatsApp groups auto-joined on primary model
• 🪝 Hooks (boot-md, command-logger, session-memory) on Sonnet
• 💬 Orphaned sessions still holding tokens
• 🤖 maxConcurrent too high — burst cost multiplier
• 📁 Workspace root bloat inflating context
• ⚙️ Primary model cost awareness

All findings include exact fix commands.