[Feature Request] Add `--advisor` flag to Claude Code CLI

Summary

Add an --advisor <model> flag to the claude CLI (and to claude agents) so an advisor can be attached at launch/dispatch time, matching the existing --model and --effort flags.

Motivation

The advisor is reachable today via the interactive /advisor command and the advisorModel setting. Neither cleanly attaches an advisor to a non-interactive or dispatched session: claude agents (background sessions) accepts --model and --effort but has no --advisor, and background subagents don't inherit the parent's advisor (#76381).

This blocks a common cost-optimization pattern — pairing a cheaper executor (Sonnet/Haiku) with a stronger advisor (Opus) that reviews only at decision boundaries ("buy the ceiling before the floor"). A strong advisor lets you run the executor at a lower tier/effort, which is often cheaper than raising the executor's own tier for equivalent quality.

Concrete example

An autonomous session planned a batch of backlog tasks against a fixed budget and wanted to dispatch each to a Sonnet worker with an Opus advisor attached. With no --advisor flag on claude agents, there was no way to do that at dispatch — the fallback was to run a stronger executor and review manually, exactly the cost the advisor pattern is meant to avoid.

Proposed shape

claude --advisor opus
claude agents --advisor opus --model sonnet --effort medium

Same grammar as the existing --model / --effort flags; --advisor sets the session's initial advisorModel binding.

Related

• #73072 — per-subagent advisor frontmatter field (config-file counterpart of this request)
• #76381 — background subagents don't inherit the advisor (the gap this flag would close for dispatched sessions)
• #78034 — advisor auto mode for autonomous operations (adjacent)

---
Environment: Platform linux · Terminal cursor · Claude Code 2.1.211 · Feedback ID 27449b4f-97cd-4795-83f9-e69384f82486
(Originally filed via voice feedback; rewritten as a structured request.)