Model Pilot v2.0
Install: clawhub install model-pilot
Stop overpaying for intelligence. Match every task to the cheapest model that delivers the quality you need.
Language
Detect from user's message language. Default: English.
When to Activate
- User explicitly asks about model choice or cost
- A task is starting that will be expensive (agent judgment — don't analyze trivial tasks)
- User says "save tokens", "cheaper model", "right model?"
Do NOT activate for trivial tasks — formatting, greetings, simple lookups.
Quick Complexity Check
Before an expensive task, ask yourself:
1. What is the task? (classify)
2. What model is active?
3. Could a cheaper model handle this equally well?
4. If yes → recommend switch + estimated savings
5. If no → proceed, explain why
Tier Classification
| Tier | Tasks | Model |
|---|---|---|
| 🟢 Routine | Formatting, Q&A, reminders, greetings, file ops | Cheapest available |
| 🟡 Intermediate | Email drafting, code review, data analysis, translations | Mid-tier |
| 🔴 Complex | Architecture decisions, creative writing, debugging complex issues | Best available |
Signal: Task needs <1000 output tokens + no creativity = Tier 1.
Live Pricing
Fetch current pricing when user asks about costs:
web_fetch https://z.ai/pricing
web_search "openai gpt-4o pricing per token"
Do NOT hardcode prices — they change frequently. Always fetch live.
Cost Estimation
Quick formula:
Estimated cost = (input_tokens / 1M × input_price) + (output_tokens / 1M × output_price)
Show to user:
Model: GLM-5-Turbo
Input: ~5K tokens
Output: ~2K tokens
Estimated cost: ~$0.002
Could use: GLM-4 (~$0.001) — saves ~50%
Quick Commands
| User says | Action |
|---|---|
| "right model?" | Quick complexity check |
| "model cost" | Cost estimate for current task |
| "compare models" | Live pricing comparison |
| "cheaper option" | Find cheaper model for same task |
Guidelines for Agent
- Don't waste tokens analyzing cheap tasks — check only for expensive ones
- Use live pricing — never hardcode
- Be honest about quality — if cheaper means worse, say so
- Keep it quick — 5 seconds of thinking max
- Match user language
What This Skill Does NOT Do
- Does NOT switch models automatically — recommends only
- Does NOT persist anything
- Does NOT access credentials
- Does NOT modify any files
More by TommoT2
- cross-check — Auto-detect and verify assumptions in your responses
- context-brief — Persistent context survival across sessions
- setup-doctor — Diagnose and fix OpenClaw setup issues
Install the full suite:
clawhub install model-pilot cross-check context-brief setup-doctor