fable5

Long Horizon Terminal Benchmark with Dense Reward Grading

Opus에 근접한 품질을 Sonnet 단가에 — 가성비 AI 운영 모델을 직접 구축한 Claude Code 프로젝트

Make Opus 4.8 (or any Claude model) work like Claude Fable 5 — a Claude Code skill + guard hooks (plan gate, model ceiling, per-task enforcement) for Fable-5-grade discipline: plan, self-verify, and route sub-agents, without Fable 5.

Proactive 5h usage guard for Claude Code (macOS). Pause before extra wallet on long batch/subagent runs. Account-level window visibility. /usage-guard skill + daemon. Not affiliated with Anthropic.

Reverse-engineer the RL reward an LLM was trained on — from black-box agentic coding behavior alone. Tested on Opus 4.8, Fable 5, GPT-5.5.

After distilling Fable 5, the output is an indexed CLAUDE.md + on-demand rules + skills, an executable specification designed for weaker model sessions.

In Jiorimetry, every redstone circuit is derived from the game's source code, requirements, and constraints.

Zero-dependency watcher that notifies you the moment Claude Fable 5 is reachable again in Claude Code (Telegram/Discord/ntfy/desktop). MIT.

受 Fable 5 启发、面向 Codex CLI 与桌面版的工作纪律 Skill、Hooks 和严格任务编排器。

Universal AI skill: converts messy chat transcripts into rigorous /goal loop prompts for agentic models like Anthropic Fable 5.

116-check security, bug, and data-protection audit skill for Claude Code — with live threat intel, 0-100 scoring, and vibe-coding-specific patterns

Claude Fable5 distillation data and a reproducible 7B SFT workflow.

VoxelCraft is a high-performance voxel game framework written entirely in pure Python and generated from scratch by Claude Fable 5. Inspired by Minecraft, it leverages numpy vectorization for fast chunk updates, features rich physics, flowing water, day-night cycles, villages, and mobs — all running smoothly at 60 FPS.

A ready-to-run agentic coding setup around TeichAI/Gemma-4-31B-Fable-5-Agent-Distill (vision + thinking + tools), with an agentic work method baked into the model's system prompt and a single-file tool-executor harness. Runs on ~24 GB VRAM at Q6_K (smaller quants work with less).