Fable Prompt Skill

A Claude Code skill that turns a rough, half-formed idea into a clarified plan — and, once approved, a working implementation.

It packages an “unknowns-first” workflow: instead of guessing at what you meant and diving straight into code, the skill interviews you one question at a time, scans the codebase for hidden constraints, brainstorms or prototypes when the solution space is unclear, produces an approval-gated implementation plan, and finishes with a plain-English explainer and quiz so you can sanity-check the change before it ships.

What’s in this repo

Inside the zip:

fable-prompt/
├── SKILL.md                          # The skill definition Claude Code reads
├── README.md                         # Skill-local install notes
├── TRIGGER.md                        # How the /fable-prompt command is triggered
├── templates/
│   ├── implementation-notes.md       # Template used while implementing
│   └── explainer-and-quiz.md         # Template used for the post-build report
└── reference/
    └── question-bank.md              # Bank of discovery-interview questions

What it does

When you invoke /fable-prompt, Claude Code walks through up to six phases:

1. Discovery interview — asks one high-leverage question at a time (goal, constraints, taste, autonomy level) instead of dumping a questionnaire on you.
2. Blind-spot pass — inspects the repo, docs, tests, and configs to surface hidden constraints, risky assumptions, and likely failure modes before proposing a plan.
3. Brainstorm / prototype — when the right approach isn’t obvious, it generates 5–10 ranked approaches, or builds a quick throwaway prototype (e.g. a single HTML mockup) so you can react to something concrete instead of describing taste in words.
4. Implementation plan — a structured plan that leads with the decisions you actually need to weigh in on (data model, UX, edge cases, verification), with an explicit approval gate. Nothing gets built until you approve it or explicitly say to proceed.
5. Implementation — builds only the approved plan, keeps changes scoped, logs any deviations and open questions, and stops to ask if it hits a high-risk unknown.
6. Explainer + quiz — after the change lands, produces a reviewer-friendly summary (what changed, why, files touched, known limitations, follow-ups) plus a short quiz to confirm you actually understand the change before merging.

The core idea: the prompt is only the map — the real work is the territory. The skill’s job is to close the gap between what you asked for and what actually needs to be true, by surfacing known unknowns and unknown unknowns instead of silently assuming.

Installation

Option A — Project-only (recommended for a single repo)

unzip fable-prompt-skill-v2.zip
mkdir -p .claude/skills
cp -R fable-prompt .claude/skills/fable-prompt

Option B — Personal (available in every project)

unzip fable-prompt-skill-v2.zip
mkdir -p ~/.claude/skills
cp -R fable-prompt ~/.claude/skills/fable-prompt

Option C — Upload to claude.ai

If your plan supports custom Skills on claude.ai, upload fable-prompt-skill-v2.zip directly (or re-zip the extracted fable-prompt/ folder) in Claude’s Skills settings.

Usage

Once installed, restart Claude Code (or start a new session) and run:

/fable-prompt

This starts the discovery interview from scratch.

You can also skip straight to a phase, or seed it with a task:

Example:

/fable-prompt Build a new onboarding dashboard and help me discover the unknowns first

What to expect

• Claude asks one question at a time and waits for your answer — it won’t front-load a wall of questions.
• It will not implement anything until you approve a plan or explicitly tell it to proceed.
• A visible Unknowns Board is maintained throughout discovery and planning so you can see what’s still uncertain.
• If it hits an unknown mid-implementation, it makes the conservative, reversible choice and logs it — or stops and asks if the choice is risky/irreversible.

When to use it

Reach for /fable-prompt at the start of any ambiguous piece of work — a new feature, a fuzzy bug report, a “make it feel more premium” design ask — where jumping straight to code risks building the wrong thing. Skip it for small, well-specified changes where the ask is already unambiguous.

License

No license specified — all rights reserved by the author unless stated otherwise.