Advisor

Use the advisor tool to get a stronger model's second opinion at key decision points during long Claude Code tasks.

What Advisor Does

Advisor lets Claude Code consult a stronger or separate model at key moments. It is not another worker editing files. It is a second-opinion tool for planning, debugging, and deciding whether a task is truly complete.

Use advisor when most turns are routine but the quality of a few decisions matters:

• Large refactors with several plausible architectures.
• Debugging loops where the same error keeps returning.
• Risky migrations where completion needs independent scrutiny.
• Final checks before declaring a goal complete.

Enable Advisor

Use the slash command:

/advisor opus

Or launch a single session with:

claude --advisor opus

You can also configure a persistent default in settings:

{
  "advisorModel": "opus"
}

The advisor must be at least as capable as the main model. Availability depends on your Claude Code version, model, and account access.

Advisor vs Other Ways to Get a Second Opinion

Advisor is lightweight compared with a team or workflow. It is best when the hard part is judgment, not parallel labor.

Good Advisor Prompts

Ask Claude to consult the advisor when the decision is crisp:

Before implementing, ask the advisor to challenge this migration plan.
Focus on rollback risk, data consistency, and hidden dependencies.

Or before finishing:

Ask the advisor whether the goal is actually complete.
Use the diff, test output, and remaining TODOs as evidence.

When Not to Use Advisor

Skip advisor when:

• The task is small and obvious.
• You need many workers, not one judgment call.
• Every turn needs maximum reasoning, in which case switching the main model is simpler.
• You need deterministic enforcement, where hooks or tests are better.

Official References

• Advisor↗
• Model configuration↗
• Dynamic workflows↗
• Subagents↗