Give each subagent its own effort: low for exploring, high for verifying

Since v2.1.292 (October 6), Claude can pass an effort when it spawns a subagent. The effort field in a definition file existed before. Together they let you search cheaply and verify expensively.

<!-- .claude/agents/verifier.md -->
---
name: verifier
description: Re-checks a finished change against the plan and the tests. Use before declaring done.
tools: Read, Grep, Glob, Bash
model: sonnet
effort: high
---
Run the checks, compare with PLAN.md, report only gaps that affect correctness.

You can also just say it in the main session: "run the verifier subagent at high effort". The session default changes with /effort.

The claude.dev post has the rule of thumb for picking a level: low for brainstorming and easy changes, medium for ordinary feature work, high for edge-case-heavy work like bugs in an existing codebase, max for unattended autonomous runs. Higher effort catches more edge cases; it does not fix a wrong approach.