pick-modellisted

# Picking the best model for the task Match the task to the cheapest model that will still do it well. Spending Opus on a mechanical rename wastes money and time; spending Haiku on an architecture decision produces a confident wrong answer. This skill picks the tier (and effort) and then acts on it. ## How model selection actually works here (read first) You **cannot** switch the main session's model yourself — `/model` is a user action, and hooks can't route models. The one lever you control is **subagents**: dispatch a self-contained sub-task to a model-pinned subagent via the Task tool. So there are exactly two moves: - **Delegate** — the work is a self-contained sub-task → dispatch the matching subagent (`quick` / `standard` / `deep`). This genuinely runs that sub-task on the chosen model. - **Advise** — the *whole session's* work belongs on a different tier → tell the user which `/model <tier>` to run and why. You can't do it for them. Default to **delegate** when the task is separable; **advise** when it defines the session. ## The tiers Always use the **aliases** `haiku` / `sonnet` / `opus` (and subagents `quick` / `standard` / `deep`) — an alias always resolves to the *latest* model in its tier, so you never pin a stale version. The "Currently" column is informational and will drift; the routing logic never depends on the exact version. | Tier | alias · subagent | Currently | Relative cost | Use for | |---|---|---|---|---| | Quick | `haiku` · `quick` | Hai