fable-advisor — install & setup

This skill approximates the Claude API "advisor tool" (which is API-only and not available in Claude Code) by spawning a Fable 5 subagent as a second opinion at genuine decision points. The working split: Opus orchestrates, Sonnet does workers, Fable advises.

Two-part design — read this before installing on a new machine

The advisor is made of two pieces, and only one of them lives in this folder:

1. SKILL.md (in this folder) — the consultation procedure: when to consult, how to package context (focused brief vs. read-only investigation), and how to weight the advice. This gives you on-demand use: invoke it with /fable-advisor.
2. A rule in ~/.claude/CLAUDE.md (NOT in this folder) — the standing instruction that makes the advisor fire automatically at decision points. A skill can't self-trigger on that kind of always-on meta-behavior, so the automatic half must live in your global CLAUDE.md.

Consequence: installing this skill on a new machine gives you the on-demand skill, but the automatic firing does NOT travel with it — you must also add the CLAUDE.md rule below.

Enable automatic firing (paste into ~/.claude/CLAUDE.md)

## Fable Advisor — auto-consult at decision points (OPUS-MAIN sessions ONLY)

Working setup: **Opus orchestrates, Sonnet does worker subagents, Fable advises.** The Claude
API "advisor tool" is API-only and NOT available in Claude Code, so the advisor is a
**Fable 5 subagent** driven by the `fable-advisor` skill.

**ONLY when the session's main model is Opus, proactively consult a Fable advisory subagent at
genuine decision points** — before committing to a non-obvious approach, when stuck, when
weighing a change of approach, and before declaring a substantial task done (make the deliverable
durable first). Invoke the **`fable-advisor`** skill for the procedure (focused-brief vs
read-only-investigation context modes, and how to weight the advice).

**Do NOT invoke it unless main is Opus.** If main is already Fable, Fable→Fable is pure waste
(nothing stronger to consult); if main is Sonnet/Haiku, this configured path doesn't apply.
Check the actual current model first.

**Restraint — rationing is the goal:** Fable is the priciest tier and drains the shared usage
pool fastest. Reserve consults for **high-value decisions on non-trivial tasks** — NOT
routine/reactive turns, trivial edits, or simple lookups; roughly one consult before committing
to an approach and one before declaring done.

Optional companion rule (if you also run the Sonnet-workers convention): in your CLAUDE.md's subagent-model policy, note that the Fable advisor is the one sanctioned non-Sonnet subagent (spawned with model: "fable"), so it doesn't conflict with an "always spawn workers on Sonnet" default.

Install the skill itself

• Claude Code / Desktop Code mode (filesystem): put the fable-advisor/ folder in ~/.claude/skills/ (or in a git-tracked skills dir that ~/.claude/skills symlinks to). Claude Code discovers it at session start.
• From the .skill package: import fable-advisor.skill through Claude Desktop's Skills feature (opening the file surfaces a "Save skill" card).
• With just the skill installed, use it on demand via /fable-advisor. Add the CLAUDE.md rule above to get automatic firing.

Requirements

• Opus-main only. The skill applies only when the session's main model is Opus — Fable advising Fable is waste, and the setup runs Opus as the orchestrator. (See the precondition in SKILL.md.)
• Fable access on your account/plan, and the Anthropic API — the advisor/Fable tier is not available via Bedrock, Google Cloud, or Foundry.