The advisor tool in Claude Code, how it works… Bookmark this for later
The common setup is Sonnet as the executor, with Haiku for subagents or simple skills
The advisor pairs Opus on top of either one
Your main model runs the whole task. When it hits a decision it can’t resolve alone, it consults Opus for guidance, then keeps going
You stay at executor cost most of the time and only pay Opus rates on the calls that actually need it
Enable it with /advisor
or set a persistent default in settings.json:
"advisorModel": "claude-opus-4-8"
What happens under the hood when it fires:
The executor emits the advisor call with empty input, It only signals timing
The server forwards your full transcript (system prompt, every tool call, every result) to Opus on a separate inference pass
Opus runs with no tools and no context management, Its thinking is dropped
Only the advice text comes back, injected as an advisor_tool_result the executor reads before continuing
All of it inside a single request 👇
The executor emits the advisor call with empty input, It only signals timing
The server forwards your full transcript (system prompt, every tool call, every result) to Opus on a separate inference pass
Opus runs with no tools and no context management, Its thinking is dropped
Only the advice text comes back, injected as an advisor_tool_result the executor reads before continuing
All of it inside a single request 👇
Generated by Thread Navigator
Press ⌘ + S to quick-export