Smart Cost Tracker

Track AI agent spending in real time. Shows cost per message, per conversation, per day. Budget alerts, daily/weekly reports, cost-per-task breakdown. Use wh...

Security Scan

The skill is internally consistent: it's an instruction-only cost-tracking helper that stores estimates locally and does not request extra credentials or installs.

Purpose & Capability

Name and description match the instructions: the skill calculates/estimates per-message and aggregated costs and writes a local cost log. It does not request unrelated credentials, binaries, or installs.

Instruction Scope

Instructions stay within the declared purpose: estimate tokens, compute costs, and maintain ~/.openclaw/cost-log.json. The skill necessarily needs access to conversation prompts/responses to estimate token counts; the SKILL.md does not instruct storing full message text (log schema shows metadata only), but implementers could store more—users should confirm the runtime will not persist full messages.

Install Mechanism

No install spec or code files are present (instruction-only). Nothing is downloaded or written to disk by an installer beyond the described cost-log file created at runtime.

Credentials

No environment variables or credentials are requested, which is proportional. The skill writes to the user's home directory (~/.openclaw/cost-log.json) and needs access to conversation text to estimate tokens — both are expected for this purpose. Users should be aware that the log may contain task names/session metadata that could be sensitive.

Persistence & Privilege

always is false and the skill is user-invocable; it only persists its own local cost-log. It does not request system-wide config changes or elevated privileges.

Assessment

This skill appears to be what it claims: a local cost estimator that writes a JSON log at ~/.openclaw/cost-log.json and needs access to the conversation text to estimate token counts. Before installing or enabling: (1) confirm the runtime will not persist full message text (only metadata) if you care about privacy; (2) check file permissions on ~/.openclaw/cost-log.json or store it on encrypted disk if your machine is shared; (3) remember pricing is an estimate—update the model pricing in the SKILL.md if your provider's rates differ; (4) because it's instruction-only, review your agent runtime to ensure no other skill or component can read/forward the cost-log (the SKILL.md itself does not send data externally). If any of these are unacceptable, disable or do not enable tracking.

You track AI spending so the user never gets surprised by their API bill.

Core Behavior

When the user enables cost tracking, calculate the estimated cost after each response and append it to the cost log. The user controls when tracking is active. When the user asks about spending, present the data clearly.

Data Storage Disclosure

This skill writes to ~/.openclaw/cost-log.json on your local machine to track spending history. No data is sent to external services. You can delete this file at any time to clear all history. Tracking only runs when you explicitly enable it.

How to Calculate Cost

Token Pricing (as of March 2026)

Claude Opus 4.6:    $15.00 / 1M input,  $75.00 / 1M output
Claude Sonnet 4.6:   $3.00 / 1M input,  $15.00 / 1M output
Claude Haiku 4.5:    $0.80 / 1M input,   $4.00 / 1M output
GPT-5.4:             $2.50 / 1M input,  $10.00 / 1M output
GPT-5.2:             $1.50 / 1M input,   $6.00 / 1M output
GPT-5.1:             $0.60 / 1M input,   $2.40 / 1M output
DeepSeek V3:         $0.27 / 1M input,   $1.10 / 1M output

Estimation Method

If exact token counts aren't available:

• Average English word = ~1.3 tokens
• Count words in the prompt + response, multiply by 1.3
• Apply the pricing for the current model
• Round to 4 decimal places

Cost Log

Maintain a file ~/.openclaw/cost-log.json with this structure:

{
  "entries": [
    {
      "timestamp": "2026-03-27T14:30:00Z",
      "model": "openai-codex/gpt-5.4",
      "tokens_in": 1200,
      "tokens_out": 800,
      "cost_usd": 0.0110,
      "task": "summarized a paper",
      "session": "main"
    }
  ],
  "daily_totals": {
    "2026-03-27": { "cost": 0.43, "messages": 12, "tokens": 45000 }
  },
  "budget": {
    "daily_limit": 5.00,
    "monthly_limit": 50.00,
    "alert_threshold": 0.80
  }
}

Commands

"Start tracking" / "Enable cost tracking"

Begin logging costs after each message. Show confirmation: "Cost tracking enabled. I'll log token usage after each interaction."

"Stop tracking" / "Disable cost tracking"

Stop logging. Show confirmation: "Cost tracking paused."

"How much did that cost?"

Show the cost of the last message:

Last message: ~$0.0110 (1,200 in / 800 out tokens, gpt-5.4)

"Show my spending" / "cost report"

Show a summary:

# Cost Report — March 27, 2026

## Today
Messages: 12 | Tokens: 45,000 | Cost: $0.43

## This Week
Messages: 67 | Tokens: 312,000 | Cost: $2.87

## This Month
Messages: 234 | Tokens: 1.2M | Cost: $11.43

## By Model
gpt-5.4:     $8.20 (72% of spend)
gpt-5.1:     $2.15 (19%)
haiku:        $1.08 (9%)

## Budget
Daily limit: $5.00 — used $0.43 (8.6%)
Monthly limit: $50.00 — used $11.43 (22.9%)

"Set budget $X/day" or "Set budget $X/month"

Update the budget limits in cost-log.json.

"Most expensive task"

Find the highest-cost entry and show it:

Most expensive: "Generated architecture diagram" — $0.89 (42K tokens, gpt-5.4)

"Cost trend"

Show daily costs for the last 7 days:

Mar 21: $1.20 ████████
Mar 22: $0.85 █████
Mar 23: $2.10 █████████████
Mar 24: $0.43 ██
Mar 25: $1.67 ██████████
Mar 26: $0.92 █████
Mar 27: $0.43 ██  (so far)

"Always show cost on/off"

Toggle whether cost is printed after every message:

[gpt-5.4 | 1.2K in / 800 out | $0.011]

Budget Alerts

When spending exceeds the alert threshold (default 80%):

• Daily: "Budget alert: You've spent $4.20 of your $5.00 daily limit (84%)"
• Monthly: "Budget alert: You've spent $42.50 of your $50.00 monthly limit (85%)"

When over budget:

• "OVER BUDGET: You've spent $5.43 today, $0.43 over your $5.00 daily limit"

Rules

• Never block the user from working even if over budget — just warn
• Costs are ESTIMATES unless exact token counts are available
• Always say "estimated" when using word-count approximation
• Store all data locally in ~/.openclaw/cost-log.json — never send to external services
• If the cost log doesn't exist, create it on first use