ClosedLoop.ai
Resources

PRD Template

A reusable PRD template that produces artifacts agents and reviewers can execute against reliably.

Strong PRDs reduce ambiguity before a loop ever starts. They are also what the four PRD judges (prd-auditor, prd-dependency-judge, prd-scope-judge, prd-testability-judge) evaluate against.

Inputs you need

  • problem statement
  • user impact
  • constraints (tech, policy, business)
  • measurable success criteria
  • known risks

Template

# <Feature name>

## Problem
What's broken, missing, or underserved today? Keep it one or two paragraphs.

## Desired Outcome
What does the world look like once this ships? State the behavior change, not the implementation.

## User Stories
- As a <role>, I want to <action> so that <outcome>.
- As a <role>, I want to <action> so that <outcome>.

Keep this list at or below eight stories per PRD. The scope judge enforces this.

## Constraints
- Technical constraints (existing stack, API contracts, migration budgets)
- Policy / compliance constraints
- Business constraints (deadlines, cost ceilings)

## Success Criteria
Each criterion must be independently measurable.
- Metric: <what you will measure>. Target: <threshold>.
- Behavior: <observable behavior>. Target: <true/false>.

## Out of Scope
What you are explicitly deciding not to build in this PRD.

## Dependencies
- Upstream: <other work that must land first>
- Downstream: <work that will depend on this>
- External: <third-party services, APIs, vendors>

## Risks and Mitigations
- Risk: <summary>. Mitigation: <plan>.

## Kill Criteria
What would cause you to stop this work mid-flight?

Why this shape

Every section maps to one or more judges:

  • Problem + Desired Outcomeprd-auditor checks structural completeness.
  • Success Criteriaprd-testability-judge checks that every criterion is measurable.
  • User Storiesprd-scope-judge enforces the ≤ 8 stories check.
  • Dependenciesprd-dependency-judge checks for integration risk.

If your PRD passes all four, you will start planning from an artifact that the orchestrator can faithfully execute against.

On this page