write-a-prd

name	write-a-prd
description	Create thorough PRDs through repo analysis and deep user interviews. Use for "create a PRD", "write requirements", "product spec", or "implementation plan". Writes PRDs to docs/prds and can optionally decompose them into local issue markdown files.
allowed-tools	Read, Write, Edit, Glob, Grep, AskUserQuestion, Skill

Write a PRD

Produce high-quality PRDs with explicit decisions, test strategy, and clear scope.

Arguments

$ARGUMENTS may include:

• A feature/topic description
• A path to an existing spec (example: docs/specs/feature-x.md)
• Optional mode hint (docs, issues, or both)

If mode is not explicit:

• Default to docs
• Use issues when user asks for issue decomposition, tracer bullets, or backlog generation

Output Convention

Docs PRD Mode (docs, default)

• Write to docs/prds/.
• File name must match PRD topic in kebab-case:
  • docs/prds/<topic>.md
  • Example: docs/prds/homelab-service-health-dashboard.md
• If file exists, append numeric suffix:
  • docs/prds/<topic>-2.md, then -3, etc.

Issue Mode (issues)

• Generate local issue markdown files beside the PRD:
  • docs/prds/<topic>.issues/*.md
• Keep docs PRD as source of truth first, then derive issues from it.

Workflow

1. Gather Problem Context

Ask for a long, detailed problem statement:

• What problem exists
• Who is impacted
• Why it matters now
• Any solution ideas already considered

If detail is thin, ask for concrete examples and failure cases.

2. Explore Repository Reality

Inspect the codebase before drafting:

• Existing modules and architecture
• Similar features and conventions
• Technical constraints and integration points
• Existing testing patterns (prior art)

Call out mismatches between assumptions and repo reality.

3. Interview Relentlessly

Drive iterative questioning until major ambiguity is eliminated.

Cover each branch of the design tree:

• Scope and explicit out-of-scope
• Actors and workflows
• Edge cases and failure recovery
• Data models, contracts, and state transitions
• Performance, security, and operational constraints
• UX states where relevant (loading, empty, error)

Do not stop at vague answers, force concrete decisions.

4. Propose Deep Modules and Test Scope

Sketch major modules to build/modify.

Prefer deep modules:

• Simple/stable interface
• Complex internals hidden behind that interface
• High testability with behavior-focused tests

Confirm with user:

• Module boundaries
• Which modules must be tested
• Critical behaviors each tested module must protect

5. Draft Docs PRD

Write markdown PRD with this template:

## Problem Statement

The problem that the user is facing, from the user's perspective.

## Solution

The solution to the problem, from the user's perspective.

## User Stories

A LONG, numbered list of user stories. Each user story should be in the format of:

1. As an <actor>, I want a <feature>, so that <benefit>

Example:
1. As a mobile bank customer, I want to see balance on my accounts, so that I can make better informed decisions about my spending

This list of user stories should be extremely extensive and cover all aspects of the feature.

## Implementation Decisions

A list of implementation decisions that were made. This can include:

- The modules that will be built/modified
- The interfaces of those modules that will be modified
- Technical clarifications from the developer
- Architectural decisions
- Schema changes
- API contracts
- Specific interactions

Do NOT include specific file paths or code snippets. They may end up being outdated very quickly.

## Testing Decisions

A list of testing decisions that were made. Include:

- A description of what makes a good test (only test external behavior, not implementation details)
- Which modules will be tested
- Prior art for the tests (i.e. similar types of tests in the codebase)

## Out of Scope

A description of the things that are out of scope for this PRD.

## Further Notes

Any further notes about the feature.

6. Optional Issue Decomposition (issues or both)

Invoke prd-to-issues to decompose the approved PRD into tracer-bullet issue markdown files.

Requirements for generated issues:

• Vertical slices, end-to-end behavior, not horizontal layer chunks
• Explicit dependencies (Blocked by) in dependency order
• Status: pending metadata for workflow tracking
• Acceptance criteria checklists that are externally verifiable

7. Validate Outputs

For docs PRD:

• Check sections complete and coherent
• Check user stories cover main + edge workflows
• Check implementation/testing decisions are concrete and behavior-focused

For issue output:

• Check first issue is smallest end-to-end tracer bullet
• Check dependency graph is valid
• Check each issue has status, blockers, and acceptance criteria

8. Handoff

Report:

• Output file paths
• Key decisions made
• Remaining open questions/TODOs