advisor
Advisor
Summon a top-tier advisor to think hard about one focused question, then apply its guidance and keep working. This is the escalation pattern: a cheaper model drives the main loop and consults a stronger model at the decisions that determine the outcome.
You, the calling agent, stay in charge. The advisor gives an opinion; you decide. Any agent or subagent can invoke this.
The advisor is briefed, not shown the conversation. A scoped briefing beats dumping the full transcript: it is cheaper, it does not anchor the advisor to your existing reasoning traces, and it keeps its attention on the core question.
When to use
- A hard design or architecture decision with real tradeoffs.
- A bug or failure that keeps recurring after reasonable attempts.
- A sanity check before committing to a risky or expensive approach.
- A final review before declaring a hard task done.
Reach for it proactively at these points, not only when the user asks. Skip it for routine work, quick lookups, or anything where the answer is already clear. It adds little when every step is easy and it costs a top-tier model call. For a multi-model council rather than one advisor, use moa or peer-review instead.