Finding-Your-Unknowns Claude Code Guidelines

Maintained by 泊舟 Bozhou — I write about AI and agentic coding on X, follow me at x.com/bozhou_ai.

English | 简体中文

A single CLAUDE.md file that turns Thariq's (Anthropic) field guide — A Field Guide to Fable: Finding Your Unknowns — into behavioral guidelines for Claude Code.

The Problem

From Thariq's article:

Working with Claude Fable 5 keeps re-teaching me an old lesson: the map is not the territory.

The difference between the map and the territory is what I call unknowns. When Claude runs into an unknown, it needs to make a decision based on its best guess of what I want.

Fable is the first model where I find the quality of the work is bottlenecked by my ability to clarify its unknowns.

Your prompt is a map. The codebase is the territory. Everything you didn't say — and everything you didn't know to say — is an unknown, and every unknown forces Claude to guess. Accumulated wrong guesses are how long tasks go badly off course.

The Solution

Make Claude an active partner in discovering your unknowns, at every stage:

The Four Kinds of Unknowns

• Known knowns — what's in your prompt.
• Known unknowns — what you know you haven't figured out yet.
• Unknown knowns — what's so obvious you'd never write it down, but you'd recognize it if you saw it.
• Unknown unknowns — what you haven't considered at all.

The guidelines instruct Claude to surface the last three instead of silently guessing through them.

Install

Global rules (recommended) — applies to every Claude Code session:

echo "" >> ~/.claude/CLAUDE.md
curl https://raw.githubusercontent.com/bozhouDev/finding-unknowns-skills/main/CLAUDE.md >> ~/.claude/CLAUDE.md

Per-project — append to the project's CLAUDE.md:

echo "" >> CLAUDE.md
curl https://raw.githubusercontent.com/bozhouDev/finding-unknowns-skills/main/CLAUDE.md >> CLAUDE.md

Key Insight

From Thariq:

Every explainer, brainstorm, interview, prototype, and reference is a cheap way to find out what you didn't know before it gets expensive to fix.

When a long-horizon task comes back wrong, it's usually not the model — it's undefined unknowns. Don't just retry; go back and define them.

How to Know It's Working

• Claude asks one architecture-changing question instead of five trivial ones.
• Claude offers mock prototypes before touching real code when the direction is fuzzy.
• Deviations from the plan show up in implementation-notes.md instead of silently in the diff.
• You can pass the quiz about a big change before you merge it.

Tradeoff Note

These guidelines bias toward discovery over speed. For trivial tasks (typo-level fixes), Claude is told to use judgment and skip the ceremony.

Credits

• Original methodology: Thariq (@trq212), Anthropic — A Field Guide to Fable: Finding Your Unknowns
• Repo structure inspired by andrej-karpathy-skills
• Curated and maintained by 泊舟 Bozhou

License

MIT