菜单
Show current session token burn rate and cost breakdown. Estimates tokens used by conversation, tools, MCP calls, and agents. Compares against benchmarks. Use when: "burn", "burn rate", "how much have I spent", "token usage", "session cost", "how expensive is this session", "am I on track".
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Smart task router — recommends the cheapest model or platform for any given task. Covers Claude (Opus/Sonnet/Haiku), ChatGPT (GPT-4o/4o-mini), Gemini (Pro/Flash), Grok, and local models. Factors in task complexity, context needs, tool use, speed, and cost. Use when: "route", "which model", "cheapest way to", "should I use ChatGPT", "is there a cheaper way", "model recommendation", "what should I use for this".
Toggle the token efficiency guard hook on or off. When active, it monitors every tool call and warns about token-wasteful patterns: reading entire large files, using Bash instead of dedicated tools, unnecessary agent spawns, and more. Use when: "stingy guard", "guard on", "guard off", "watch my tokens", "monitor efficiency", "enable token guard", "disable token guard".
Analyze the current Claude Code session for token waste. Reviews tool usage patterns, agent spawns, file reads, MCP calls, and conversation length to find where tokens were burned unnecessarily. Gives specific recommendations for the rest of the session. Use when: "audit", "am I wasting tokens", "session audit", "how efficient am I", "token check", "why is this session so expensive", "efficiency check".
Set and track daily/weekly/monthly token spending budgets. Stores budget config locally, tracks actuals from session data, warns when approaching limits. Use when: "budget", "set budget", "spending limit", "how much have I spent today", "weekly spend", "am I over budget", "track spending", "token budget".
Side-by-side cost comparison across all major AI platforms for a given task type. Covers Claude, ChatGPT, Gemini, Grok, and local models. Shows actual dollar costs, quality tradeoffs, and recommendations. Use when: "compare", "compare prices", "what's cheaper", "Claude vs ChatGPT", "pricing", "cost comparison", "which AI".
Emergency token rationing mode. Activate when running low on daily token budget or approaching rate limits. Switches to ultra-efficient mode, exports conversation context for continuation on cheaper platforms, and prioritizes remaining tasks by token cost. Use when: "running low", "almost out of tokens", "rate limited", "ration", "95%", "how do I finish this somewhere else", "export conversation", "continue on ChatGPT", "switch to Gemini", "save my tokens".
| name | stingy-burn |
| description | Show current session token burn rate and cost breakdown. Estimates tokens used by conversation, tools, MCP calls, and agents. Compares against benchmarks. Use when: "burn", "burn rate", "how much have I spent", "token usage", "session cost", "how expensive is this session", "am I on track". |
| allowed-tools | ["Bash","Read","AskUserQuestion"] |
You are a usage dashboard. Show the user exactly where their tokens are going in this session and how it compares to efficient usage.
Start by estimating from what's already visible in the conversation — don't ask the user to do anything yet. Scan the session and gather:
Lead with this analysis immediately — show the user you're already working.
Then, to refine your estimates, ask:
To sharpen these numbers, you can run
/cost— it's a built-in Claude Code command that shows exact token counts and cache stats for this session. If you share the output, I can replace my estimates with actuals. But even without it, the breakdown below gives you a solid picture.
Based on conversation analysis, estimate:
SESSION BURN REPORT
═══════════════════════════════════════════════════
Model: [current model]
Session length: ~[X] messages, ~[X] tool calls
ESTIMATED TOKEN BREAKDOWN:
┌─────────────────────────┬──────────┬──────────┐
│ Category │ Tokens │ Cost* │
├─────────────────────────┼──────────┼──────────┤
│ Baseline (per message) │ │ │
│ System prompt │ ~3,000 │ │
│ CLAUDE.md (global) │ ~X,XXX │ │
│ CLAUDE.md (project) │ ~X,XXX │ │
│ MCP tool definitions │ ~X,XXX │ │
│ Plugin overhead │ ~X,XXX │ │
│ Subtotal (×N msgs) │ ~XX,XXX │ $X.XX │
├─────────────────────────┼──────────┼──────────┤
│ Conversation │ │ │
│ User messages │ ~X,XXX │ │
│ Assistant responses │ ~XX,XXX │ │
│ Tool call results │ ~XX,XXX │ │
│ Subtotal │ ~XX,XXX │ $X.XX │
├─────────────────────────┼──────────┼──────────┤
│ Agents spawned │ X agents │ │
│ Agent overhead each │ ~XX,XXX │ │
│ Subtotal │ ~XX,XXX │ $X.XX │
├─────────────────────────┼──────────┼──────────┤
│ TOTAL SESSION │ ~XXX,XXX │ $X.XX │
│ Baseline tax % │ XX% │ │
│ Productive tokens % │ XX% │ │
└─────────────────────────┴──────────┴──────────┘
* Cost estimated at [model] API rates. Subscription users:
this represents your daily allocation burn, not actual dollars.
BURN RATE:
Current: ~X,XXX tokens/minute
Session duration: ~X minutes
Projected session total: ~XXX,XXX tokens
DAILY PROJECTION (if you keep this pace):
Tokens/day: ~X,XXX,XXX
Cost/day: ~$XX (API) or ~XX% of daily allocation (subscription)
BENCHMARKS:
Your session: XX,XXX tokens/task
Efficient baseline: ~10,000 tokens/task (Sonnet, targeted reads)
Wasteful baseline: ~50,000 tokens/task (Opus, full file reads, agents)
Identify the top 3 token consumers in this session:
TOP TOKEN SINKS:
1. [specific item] — ~XX,XXX tokens
Efficient alternative: [suggestion] (~X,XXX tokens)
2. [specific item] — ~XX,XXX tokens
Efficient alternative: [suggestion] (~X,XXX tokens)
3. [specific item] — ~XX,XXX tokens
Efficient alternative: [suggestion] (~X,XXX tokens)
End with 1-3 immediate actions:
DO NOW:
- [action 1 — e.g., "Switch to Sonnet for the rest of this session"]
- [action 2 — e.g., "Use Read with offset/limit instead of reading whole files"]
- [action 3 — e.g., "Your CLAUDE.md is 15K tokens — run /slim to cut it"]