Pin a cheaper model and fork the context in your skill frontmatter

The problem

Skills inherit the session model, effort level and full conversation context. A trivial lookup skill runs on your most expensive model and drags the entire conversation along with it.

The hack

Add `model`, `effort` and `context: fork` to a skill's frontmatter. The skill runs on the model and effort you name, in a forked subagent context that does not carry your conversation.

Why it works

The override applies only while the skill is active and is not saved to settings, so the session returns to normal on your next prompt. Forking means the skill starts clean instead of paying to re-read a long conversation it does not need.

Setup

1. Open the skill's `SKILL.md`. 2. Add `model: haiku` (or any value `/model` accepts, or `inherit`). 3. Add `effort: low` for mechanical work. 4. Add `context: fork` if the skill does not need your conversation history. 5. Invoke the skill and confirm behaviour is still correct at the lower setting. 6. Audit your other skills the same way — bias toward cheaper for anything that is lookup, formatting or validation.

Code

---
name: verify-resource
description: Check that a documentation link resolves and summarize the page in one line.
model: haiku
effort: low
context: fork
---

Fetch the URL, confirm it returns 200, and report the page title.

Additional details

Source

Related

• → Turn recurring prompts into skills you invoke with /name
• → Give custom subagents isolation: worktree for safe parallel edits
• → Press t in /skills to sort by token count and demote the bloat to name-only