Fable Skill — AI skill for Claude Code

fable-skill

[](https://github.com/almutaz9000/fable-skill/actions/workflows/ci.yml)

**Fable-model operating discipline for any AI coding agent.**

One skill that upgrades how your agent *works* — not what it knows. It enforces the working discipline that separates top-tier agentic models from ordinary runs: explore before planning, plan before acting, verify every change with evidence, debug at root cause instead of retrying blindly, persist state across long tasks, and self-review before declaring done.

It also **routes the workflow to the task**: before acting, the agent classifies what the user will actually consume — working code, a diagnosed failure, a sourced answer, a report, a review verdict, a plan, a data insight, or a changed system — and runs the matching playbook (build, debug, research, write, review, plan, analyze, operate). Verification for a report means fact/structure/reader edit passes; for a data analysis it means row counts and magnitude checks; for a deploy it means dry-runs and rollback plans — not one code-shaped process forced onto everything.

Install it once with `npx`, natively, into whichever agent you use:

Claude Code	Cursor	GitHub Copilot	OpenAI Codex
Gemini CLI	Google Antigravity	Windsurf	Cline
Amp	OpenClaw / ClawBot	Hermes Agent	Aider
Continue.dev	Zed	JetBrains Junie	Kiro (AWS)
Trae	Qwen Code	OpenCode	Goose
Warp	Kilo Code	Augment	OpenHands
Replit Agent	Crush (Charm)	any AGENTS.md agent	claude.ai (zip)
any model via system prompt

And it is **model-agnostic by design**: the same discipline runs on frontier models and on free or local ones — Gemma, Qwen, Kimi K2, GLM (Z.ai), Llama, DeepSeek, Claude Haiku, Grok, GPT — with built-in fallbacks when a platform lacks parallel tool calls, subagents, or a question UI. For models with no codi