Agentic AI / Skills

Model Recommend

Skill

Recommends the cheapest Claude model that is sufficient for a task, compares it with your session's model, and tells you whether to run it here or open a new session.

Best for

  • Keeping AI costs under control
  • Routing simple work to a small model and hard work to a large one
  • Teams running several sessions at once

What you give it

  • The task or change record

What you get back

  • The recommended model with a one-line reason
  • A same/different verdict against your current model
  • The command to open a session on the right model

How it works

  1. Judges files touched, ambiguity, reasoning depth, risk and reversibility.
  2. Maps the task to the tier table below.
  3. Compares with the session's model and states the verdict.
  4. Never switches models by itself.

Example

You: /model-recommend rename a label in 14 files

Result: Haiku — mechanical, well-specified, reversible. This session is Opus: recommended cheaper model differs. Open a new session with `claude --model <haiku id>`, or say "continue here".

Limits — please read

  • A rule of thumb, not a benchmark.
  • Model names and IDs change: update the table when they do.