Parable is a community-made skill set and hook mechanism that distills the working discipline of Claude Fable 5 into a pseudo-Fable mode for Claude Opus and Sonnet, based on one user's study showing 90% of performance gap was procedural.

Stars

21

7-day growth

No data

Forks

0

Open issues

0

License

MIT

Last updated

2026-07-08

AI repository intelligence
FR-AI / ANALYSIS

Why it is worth attention

It packages an empirically tested procedure—backed by a small study with defect detection rates of 4/6 solo and 6/6 in ensemble—that aims to make Opus behave more like Fable through skill files, hooks to prevent rule decay, and a transferable testing protocol.

Who it is for

  • Claude Code users seeking Fable-like discipline without Fable access
  • Developers wanting to improve AI-assisted code audits and planning
  • Teams exploring cost-efficient workflows by using cheaper models for well-briefed tasks
  • Researchers interested in procedural scaffolding for frontier LLMs

Use cases

  • Single-agent code audits with defect recall (4/6 solo found in study)
  • Multi‑agent adjudicated ensembles for hard problems (achieved 6/6 in study)
  • Planning and iterative refinement using planning gates (P1–P9) with evidence paste
  • Cost reduction: using cheaper models in implementation seats with a stronger model review gate

Strengths

  • Evidence‑based: experiment data (solo & ensemble recall, planning brief ratchets) is published and replicable via provided testing protocol
  • Procedural focus: claims 90% of Fable's advantage is procedure, not intelligence, and provides concrete rules and hooks
  • Hook mechanism prevents rule decay in long agent runs (rule card reinjected every prompt and every 5 minutes)
  • Open source (MIT) and ready to fork—no contributions needed, just adapt placeholders

Considerations

  • Numbers come from one small study on one codebase by one person—treat as directional, not definitive
  • Coordinator layer requires Claude Code's agent/notification machinery; director + worker split works more broadly
  • Skill files cannot increase model capacity; for novel judgment calls the strongest model is still needed, and the method puts that model at the review gate

README quick start

Usage

Description

A pseudo-Fable mode for Claude Opus — Fable-style planning, verification, and multi-agent discipline as a Claude Code skill set. Unofficial; not affiliated with Anthropic.

Related repositories

Similar projects matched by category, topics, and programming language.

Vincentwei1021
Vincentwei1021 GitHub avatar

video-shotcraft

An AI agent skill that turns Claude Code or Codex into a motion-design studio for crafting cinematic product videos with Remotion, offering 106 shot recipe cards, 162 styles, 161 motion previews, and a production-ready template.

TypeScript
2,432
yanhua1010
yanhua1010 GitHub avatar

self-media-content-workflow

A modular, tool-agnostic suite of Agent skills for the complete self-media content production lifecycle, from fuzzy brief to delivery and analytics, with mandatory human-in-the-loop checkpoints.

Python
295
bjan
Featured
bjan GitHub avatar

claudemux

Claudemux is a lightweight, dependency-minimal message bus that lets multiple Claude Code sessions communicate in real-time over tmux, enabling one session to ask another a question and receive the answer without manual intervention.

AI & Machine LearningAI Agents
9