subagentprompts

.com prompt engineering
grounding: code.claude.com/docs/en/advisor.md

The advisor tool, and the pairing this session actually used

The advisor tool lets a main model consult a second, typically stronger model at key moments during a task — before committing to an approach, when stuck on a recurring error, or before declaring a task complete. The advisor receives the full conversation, including every tool call and result, and returns guidance the main model applies before continuing. It is experimental, requires Claude Code v2.1.98 or later with the Anthropic API, and is not available on Amazon Bedrock, Google Vertex AI, or Microsoft Foundry.

Accepted pairings

The advisor must be at least as capable as the main model:

Main modelAccepted advisorsNotes
Haiku 4.5Fable, Opus, SonnetHaiku can call the advisor but cannot act as one
Sonnet 4.6Fable, Opus, Sonnet
Opus 4.6 or laterFable, Opus at or above the main model's versionAn Opus 4.7 main with an Opus 4.6 advisor is rejected
Fable 5 (v2.1.170+)FableAn Opus or Sonnet advisor is rejected

What this session actually did

This session's main model is Claude Sonnet 5, running inside Cowork. Sonnet-main-plus-Fable-advisor is an explicitly accepted, documented pairing: "Fable 5 guidance at decision points without running Fable 5 throughout. Requires v2.1.170 or later and Fable 5 access." In practice, this repo used Fable 5 twice in two different roles, not one:

SiteFable 5's roleWhy that role
subagentcode.comPrimary builder — a full Agent-tool dispatch with model:"fable"The operator asked to route the whole build to Fable 5 directly.
subagentprompts.com (this site)Advisor — reviewed the drafted content and seed data before deployMatches the real advisor pattern: guidance at a decision point, not the whole task.

Both are legitimate uses of Fable 5 inside a Sonnet 5 session. Neither is Claude Mythos 5, which stays Project-Glasswing-restricted and unused anywhere in this repo — see subagentcode.com's Fable 5 page for that verification.

When to reach for the advisor instead of switching models

The advisor fits long, multi-step tasks where most turns are routine but plan quality determines the outcome — large refactors, debugging sessions where an error keeps recurring, or work you want independently checked before it's declared done. It adds less value on short tasks with little to plan, or work where every turn needs the strongest model; for those, switch the main model outright instead.

Source: code.claude.com/docs/en/advisor, also mirrored in this repo's docs/docs/code.claude.com/docs/en/advisor.md.