Prompting Claude Fable 5

Fable 5 (claude-fable-5) is Anthropic's Mythos-class model above Opus. claude-mythos-5 is the same model without safety classifiers (limited release). This skill covers only what differs from prior Claude models; standard prompt engineering still applies.

Hard constraints

• Never instruct the model to echo, transcribe, or explain its internal reasoning in response text. This triggers stop_reason: "refusal". When migrating prompts or skills, audit for "show your thinking" / "explain your reasoning step by step in your answer" language. Any reasoning surfaced should be a concise summary.
• Safety classifiers refuse on request shape, not intent.

Deprescribe when migrating

Instruction following is strong enough that one brief instruction replaces an enumerated behaviour list. Prompts and skills written for earlier models are often too prescriptive for Fable 5 and degrade its output. When migrating: delete instructions, test default behaviour, and re-add only what measurably changes it. Heavy MUSTs, exact step sequences, and case-by-case enumeration are the smell.

Failure modes and tested mitigations

Each mitigation below is a condensed version of Anthropic's tested instruction; drop the relevant one into the system prompt when the failure appears.