v1.0.0 — Grade A (100%) Release

Fable 5 Skills v1.0.0 — Grade A (100%) Release

AI reasoning skills distilled from 4,665 real Claude Fable 5 chain-of-thought traces, mathematically tuned and validated.

🎯 Highlights

• 5 skills (fable-think, fable-code, fable-debug, fable-architect, fable-verify) — one SKILL.md each, drop-in for any agent that reads markdown skills.
• 31/31 quantified claims verified against the source dataset (100% pass rate).
• 9 iterative test rounds against MiniMax M3, final score 100.00% (Grade A).
• CLI installer (npx fable5-skills init) drops the skills into the right directory for your agent in one command.
• Multi-agent support: Claude Code, Cursor, Cline, Windsurf, Continue, or generic ./skills.

📦 Install

npx fable5-skills init

Default installs into .claude/skills/ (Claude Code layout). Use --agent=cursor|cline|windsurf|continue|generic for other layouts.

🧠 The 5 Skills

Skill	Purpose	Core Loop
fable-think	Foundational per-turn reasoning	ACKNOWLEDGE → OBSERVE → EXECUTE → VERIFY
fable-code	Writing & editing code	Read → Understand → Plan → Write → Verify → Iterate
fable-debug	Root cause analysis	OBSERVE → INVESTIGATE → HYPOTHESIZE → ROOT CAUSE → FIX → VERIFY
fable-architect	System design	UNDERSTAND → DESIGN → VERTICAL SLICE → VERIFY → ITERATE
fable-verify	Quality assurance	5-phrase verification vocabulary

📊 Benchmark Results

Static skill verification (claims vs. actual data)

• Total claims verified: 31
• Passed: 31
• Failed: 0
• Pass rate: 100.00%

MiniMax M3 emulation (Round 9 — final)

Test	Score	Status
T1 reasoning	100.00% (11/11)	✅
T2 debug	100.00% (9/9)	✅
T3 code	100.00% (8/8)	✅
T4 verify	100.00% (10/10)	✅
T5 architect	100.00% (8/8)	✅
T6 self_correct	100.00% (6/6)	✅
T7 loop	100.00% (7/7)	✅
Average	100.00%	✅ Grade A

🔬 Source Dataset

Distilled from Kuberwastaken/Fable-5-traces — 4,665 rows, 60 sessions of real Claude Fable 5 chain-of-thought traces.

🚀 Quick Start

# Install into Claude Code layout (.claude/skills/)
npx fable5-skills init

# Install into Cursor layout (.cursor/skills/)
npx fable5-skills init --agent=cursor

# Install a single skill
npx fable5-skills init --only=fable-debug

# Verify package integrity
npx fable5-skills doctor

⚖️ License

MIT — see LICENSE.

🙏 Acknowledgements

• Anthropic PBC — for Claude Fable 5
• Kuberwastaken/Fable-5-traces — source dataset
• Matt Pocock — for the SKILL.md format
• MiniMax Inc. — for MiniMax M3, used as the cross-model emulation benchmark

---

npm: https://www.npmjs.com/package/fable5-skills
GitHub: https://github.com/ahmdd4vd/Fable5res