consult-fable

name

consult-fable

description

Consult Fable (claude-fable-5, the tier above Opus) as an advisor — a second opinion on a hard design decision, a stuck debugging session, or a close review call. From Claude Code, spawn a subagent on model fable; from any other harness, call the Claude Code CLI. A consultation, not an implementation hand-off. Triggers on: 'consult fable', 'ask fable', 'get fable's opinion', 'second opinion from fable', or when a judgment call is too close to make alone.

Consult Fable

Ask Fable — the frontier tier above Opus — for judgment. This is a consultation, not a delegation: Fable reads and advises, it does not edit. For implementation hand-offs, use claude-exec instead.

When to use

• A design decision with real trade-offs, where you want a frontier-tier verdict.
• A debugging dead end: present the symptoms and your hypotheses, ask what you're missing.
• A close review call (ship or hold, is this finding real, is this API shape right).

Not for lookups, mechanical work, or anything a cheaper model answers just as well — Fable is the most expensive tier (a trivial round trip already costs ~$0.5).

How to run

If you are running inside Claude Code (the Agent/Task tool is available): spawn one subagent with model: "fable" and the consultation question as its prompt. No CLI round trip needed.

Otherwise (Codex, other harnesses, plain shell): call the Claude Code CLI. Read-only tools (Read, Grep, Glob) stay available, so Fable can inspect the code it is advising on — name the relevant paths in the prompt.

claude -p --no-session-persistence --model fable --permission-mode default \
  --disallowedTools Bash,Edit,Write -- "<question>"

For long prompts, pipe via stdin and omit the prompt argument:

echo "<long question>" | claude -p --no-session-persistence --model fable \
  --permission-mode default --disallowedTools Bash,Edit,Write

Writing the question

Context is not carried over — the prompt must stand alone, in either mode:

1. Background: what is being built, and the constraints that matter.
2. The question: one decision, stated precisely.
3. The options considered, with your current lean and why.
4. Ask for a recommendation with reasoning, and what evidence would change it.

Notes

• One consultation per decision. If the answer raises a follow-up, fold it into a single second round — don't turn Fable into a chat partner.
• CLI mode: leave --effort at its default (deep reasoning is the point) and set a generous timeout (up to 10 minutes: timeout: 600000).
• The reply is advice, not an order. Weigh it against what you know, decide, and tell the user what Fable said and which parts you adopted.