Claude Prompt to Write a PRD for an AI Coding Agent

ClaudeChatGPTClaude CodeCursor🔒 Subscribers

The short answer

Interviews you, then writes a spec tight enough that Claude Code, Cursor or any agent builds the thing you meant, not the thing you said. Copy the prompt below, replace the 6 [BRACKETED] variables with your own, and paste it into Claude, ChatGPT, Claude Code or Cursor.

🔒 Subscriber prompt

You are a senior product engineer who writes specs that AI coding agents can implement correctly on the first try. Agents take everything literally, so the spec must be explicit about scope, behavior, constraints and how to verify the work.

CONTEXT
- What I want built, in my own words: [FEATURE_IDEA]
- Who uses it and why: [USERS_AND_PROBLEM]
- Existing codebase (stack, structure, conventions, relevant files): [CODEBASE_CONTEXT]
- Hard constraints (no new dependencies, must run on X, performance, security, deadlines): [CONSTRAINTS]
- Things explicitly out of scope: [OUT_OF_SCOPE]

STEP 1 - CLARIFY
Before writing anything, ask me up to 8 questions that would change the implementation. Prioritize: data model, edge cases, error states, permissions, and what "done" looks like. Wait for my answers.

STEP 2 - WRITE THE SPEC
Produce a spec with these sections:
1. Goal: one paragraph, plus a one-sentence non-goal statement.
2. User stories with acceptance criteria written as Given / When / Then.
3. Functional requirements, numbered (FR-1, FR-2...) so they can be referenced in commits and tests.
4. Data model and API or interface changes, with exact names and types.
5. Edge cases and error handling, each with expected behavior.
6. Files likely to be created or modified, and files that must NOT be touched.
7. Implementation plan as ordered milestones, each small enough to review in one sitting and ending in a runnable state.
8. Verification: tests to write, commands to run, and manual checks that prove each milestone works.
9. Open questions and assumptions.

RULES
- Be specific enough that two different agents would build roughly the same thing.
- Use [TBD] for anything I have not decided. Do not make product decisions silently; list them under assumptions.
- No vague words ("fast", "intuitive", "robust") without a measurable definition.
- Keep it under 1,200 words. Prefer lists and tables over prose.

STEP 3 - SELF-REVIEW
Reread the spec as if you were the coding agent. List any requirement you would have to guess at, fix it, and then present the final version.

This prompt is reserved for subscribers.

Pledge allegiance to the Daily Decree and the full prompt unlocks here instantly, along with every other subscriber prompt in the Vault. Unsubscribe whenever.

What to fill in

Everything in [GOLD CAPS] is yours to replace. Leave one blank and the machine will guess — and the machine guesses like a meatbag on a Monday.

[FEATURE_IDEA]
[USERS_AND_PROBLEM]
[CODEBASE_CONTEXT]
[CONSTRAINTS]
[OUT_OF_SCOPE]
[TBD]

How to use it

Actually answer the clarifying questions. Every one you skip becomes a decision the agent makes for you, silently, in a file you will not read.
Save the final spec as a markdown file in the repo (for example docs/specs/feature-name.md) and point the agent at it. Then have it reference FR numbers in its commits.
Hand the agent one milestone at a time and review in between. Even the Chosen do not merge 3,000 lines on vibes.

Example

FEATURE_IDEA: let users export their saved prompts as a CSV
CODEBASE_CONTEXT: Astro site, TypeScript, prompts stored in localStorage, no backend
CONSTRAINTS: no new npm packages, must work offline
OUT_OF_SCOPE: import, cloud sync

Works with

Tested for Claude, ChatGPT, Claude Code, Cursor. Any modern chat model that follows long instructions will cope; the weaker ones will need you to restate the output format if they drift.

Weapons from the Arsenal that pair well with this prompt:

Claude Code — Anthropic's agentic coding tool that reads your codebase, edits files and runs commands from terminal, IDE, desktop or web.
Cursor — AI-first code editor (a VS Code fork) with agents, codebase-aware chat and fast autocomplete.
Claude — Anthropic's assistant: the strongest long-form writer and careful reasoner in the consumer AI pack.

Claude Prompt to Write a PRD for an AI Coding Agent

What to fill in

How to use it

Example

Works with

More coding & automation prompts

AI Prompt for a Thorough Code Review (Bugs, Security, Readability)

AI Prompt to Explain Unfamiliar Code (at Your Level)

AI Prompt to Write and Explain a Regex (With Test Cases)

AI Prompt to Write Clear Git Commit Messages From a Diff