Fable 5 costs $50/MTok. This gives you the same thinking discipline for free.
Claude Fable 5 is Anthropic's most capable model — it plans, verifies, tests its own work, and thinks adaptively before answering. That "always-on thinking" is what makes it good.
Fable Mode is one SKILL.md file that distills that same process so any model can follow it. Haiku, GPT-4o-mini, Sonnet, whatever. Same thorough thinking. No $50/MTok bill.
Five phases the model must complete before delivering an answer:
- Understand — restate the ask, catalog constraints, detect pattern-matching traps
- Plan — decompose, generate alternatives, find the crux
- Execute — show intermediate work, tag facts KNOWN/INFERRED/UNCERTAIN
- Self-Review — verify by a different method, hunt counterexamples
- Deliver — answer first, show uncertainty honestly
Effort dial T1/T2/T3 so trivial questions skip the protocol entirely.
npx skills add overusedhydra/fable-modeOr copy SKILL.md to your agent's skills directory:
.opencode/skills/fable-mode/~/.claude/skills/fable-mode/~/.agents/skills/fable-mode/
Works with opencode, Claude Code, Cursor, Codex CLI — any agent that loads SKILL.md.
"Cheap models answer too fast. This fixes that."