A single-file operating standard for Claude Code — and most other coding agents.
FABLE.md does not make the model smarter. Capabilities live in the weights, not in prompts (that's literally section 0 of the file). What it does is encode the working discipline that separates excellent output from mediocre output on the same model:
- Honesty & pushback — "it should work" is banned: tested, or declared untested. Disagreeing with a flawed request before executing is mandatory.
- Anti-hallucination — every API/flag/version verified before it lands in delivered code; every number needs a traceable origin.
- Calibrated autonomy — a hard list of actions that always require asking first: spending money, deploys, deletions, irreversible operations, scope changes.
- Reasoning discipline — plans proportional to task size, treated as contracts; edge cases considered by default; alternatives noted for structural decisions.
- Project memory — STATUS.md updated at session end, so the next session restarts from a file, not from vibes.
- Plus: code quality rules, source hierarchy for research, communication format, and an end-of-work self-check.
-
Drop
FABLE.mdinto your project root. -
Reference it from your project's
CLAUDE.md:Read FABLE.md in full at the start of every session and treat its rules as hard constraints.Or paste its content directly into
CLAUDE.md/AGENTS.mdif you prefer a single file — the content is tool-agnostic apart from a few Claude Code conventions. -
Customize the USER PROFILE block in section 1 (name, experience, stack, communication language). This is what lets the model calibrate depth and pushback to you.
-
Priority on conflict: model safety policies > FABLE.md > project CLAUDE.md > model defaults.
- Not a jailbreak, not a "model upgrade", not magic. If a rules file promises to turn one model into a better one, it's lying to you.
- Not a new idea. Instruction files for coding agents are an established
practice (
CLAUDE.md, the openAGENTS.mdstandard). FABLE.md is a distilled, opinionated version focused on working discipline rather than project context — the two complement each other. - Not affiliated with Anthropic. Claude is Anthropic's product; the name FABLE is a nod to their model naming. This is an independent, unofficial project.
| File | Description |
|---|---|
FABLE.md |
English version |
FABLE.it.md |
Italian version (the original) |
Distilled from months of daily production use of Claude Code across web agency, hosting and product development work. First shared on r/ClaudeAI.
It's a v1.0. If a rule doesn't work in practice, is missing, or conflicts with how a specific model behaves, open an issue or a PR.
MIT — see LICENSE.
Maintained by ReviewMost4463 — GitHub.