Turn an idea into a build-ready PRD.

What is PRD Creator?

PRD Creator is a Claude Code skill that turns a raw idea into a locked-down Product Requirements Document plus build-ready milestone prompts your coding agent can execute against — one focused session at a time.

It’s built for builders who know what they want their app to do but don’t want to get stuck on data models, background jobs, or integration plumbing. The skill proposes sensible defaults in plain language and asks you to confirm or adjust.

You walk away with three things in your repo:

• _build_plan/prd.md — the locked PRD.
• milestones/N-name/prompt.md — one prompt per milestone for your agent.
• milestone-log.md — written by the agent at the end of each milestone so later sessions stay grounded.

Installation

PRD Creator ships in the open-source bm-skills marketplace. In any Claude Code session:

/plugin marketplace add buildermethods/bm-skills

/plugin install bm-prd-creator

Then open Claude Code in your project and run /bm-prd-creator, or just describe an idea you want to build — the skill auto-triggers.

Any agent

PRD Creator is just markdown — not Claude Code-specific. If you’re using Codex, Antigravity, Cursor, or another agentic tool, download the bm-prd-creator folder from the bm-skills repo and install it as a custom skill, command, or rule per your tool’s convention. The workflow is identical.

How it works

The skill runs a structured interview in a fixed sequence, locking each phase before moving on. It always proposes a default with a reason and asks you to confirm or change — editing a proposal is much easier than answering blank prompts.

You’ll move through the eight phases below. Once everything is locked, the skill writes the PRD and milestone prompts to your repo.

1. Brain dump & core purpose

Describe your idea however you want — what it is, what problem it solves, who it’s for. The skill synthesizes that into a 1–3 sentence “what we’re building” statement and asks you to confirm.

2. Features & out-of-scope

The skill proposes 4–8 core features, then proactively proposes an out-of-scope list of things v1 shouldn’t include. You confirm, cut, or pull anything back in.

The explicit out-of-scope list is one of the most valuable parts of the PRD — it stops your agent from quietly expanding the build later.

3. Tech stack & starter detection

The skill inspects your codebase — CLAUDE.md, Gemfile, package.json, folder structure — and summarizes what it finds. If you’re using the Build New starter, it factors that in so the PRD doesn’t re-spec features the starter already provides.

4. External integrations & credentials

For each feature that needs an external service (AI providers, email, storage, payments, SMS, maps), the skill proposes a default provider and lists the credentials you’ll need to obtain. The list locks before moving on.

5. Data model in plain language

The skill proposes the entities your app needs and how they relate, described in plain English (url — the link being saved, not url: string, not null). You confirm or adjust.

6. Per-feature scoping

One feature at a time, the skill locks down what’s in scope and what’s explicitly NOT — focused on what the user sees and does. Implementation choices (libraries, retries, parsing) are intentionally left to the agent during each milestone.

7. Milestone breakout

The skill proposes a milestone plan — typically 3 milestones in a sensible dependency order (Core CRUD → Integrations → Public-facing) — and offers fewer/bigger or more/smaller alternatives. Each milestone delivers something you can test in the browser.

8. The output files

Once everything’s locked, PRD Creator writes:

_build_plan/
  prd.md
  milestones/
    1-core-crud/
      prompt.md
    2-integrations-layer/
      prompt.md
    3-public-pages/
      prompt.md

prd.md is the source of truth. Each prompt.md is a thin trigger file that tells your agent to read the PRD, read prior milestone logs, plan, build, and log.

Building from the milestones

Open milestone 1’s prompt.md in your AI coding tool and ask the agent to start. The prompt directs it to read the PRD, plan in plan mode, build only what’s in scope, verify against the “Done when” criteria, and write a milestone-log.md.

When milestone 1 ships, open the next prompt and repeat.

The milestone log trail

Each milestone-log.md records what was built, decisions made, conventions introduced, and any deviations from the PRD. Later milestones read these logs alongside the PRD so they build on prior work instead of re-litigating it — this is what makes multi-session builds stay coherent.

When to reach for it

PRD Creator earns its place at the start of a brand-new build — where you’d otherwise hand your agent a fuzzy idea and let it improvise the product. It’s a strong fit when you want a clear roadmap, explicit scope boundaries, and a planning artifact your agent can reference every session.

When to skip it

Skip it if:

• The build is small enough that a paragraph in chat would do.
• You already have a planning workflow that produces scoped specs.
• You’re iterating on an existing app rather than starting something new.

Next steps

• View the source on GitHub
• Explore other Builder Methods tools