model-routerlisted
Install: claude install-skill simonlivson/claude-skills
# Model router
Pick the right tool and a model that matches the work. Start cheap, move up if the result is weak, and check the output before trusting it. Exact model names change; `/model` is the source of truth for the signed-in account.
## When to use
- Choosing a tool or model, switching models, checking usage, or handing work to another tool.
## Pick the tool
- **Claude Code:** coding and multi-file repo changes.
- **Codex CLI (ChatGPT):** a second review of code or drafts, and quick scripts.
- **Gemini CLI:** long documents, large inputs, and prompt-eval reruns.
- **ChatGPT, Gemini, or Claude web apps:** second opinions (no repo edits).
## Pick the model
| Task | Tool | Start with |
|---|---|---|
| Everyday coding, multi-file changes | Claude Code | `sonnet` |
| Simple edits, lookups, boilerplate | Claude Code | `haiku` |
| Plan, then implement | Claude Code | `opusplan` (Opus plans, Sonnet executes) |
| Hard debugging or architecture | Claude Code | `opus` (or `fable` for the longest tasks) |
| Second review, PR review, quick scripts | Codex CLI | `gpt-6.2-sol` at default or medium effort |
| Hardest review or cross-service debug | Codex CLI | `gpt-6.2-astra`; raise effort if needed |
| Small, bounded Codex edits | Codex CLI | `gpt-6-luna` at low effort |
| Long docs, large inputs, eval reruns | Gemini CLI | `auto` or `pro`; `flash` or `flash-lite` for simple or high-volume runs |
| Second opinion | ChatGPT, Gemini, or Claude web | the app default; a stronger mode