PRD Workflow

PRDs document feature requirements for AI-driven development. They provide structured context that helps AI implement features correctly across multiple files and architectural boundaries.

Quick Start

"Use the specify skill to create a PRD for [feature description]."

The specify skill guides you through an interactive PRD creation process, analyzing your codebase and proposing content for each section.

---

The SDLC Flow

For complex features, follow the full skill-driven lifecycle:

specify -> plan -> build -> review -> document

Step	Skill	What Happens
Specify	specify	AI-guided PRD creation with codebase analysis
Plan	plan	Decompose PRD into ordered tasks with architecture decisions
Build	new-feature / new-api-route / fix-bug / refactor	Implement each task with TDD
Review	code-review + design-review + qa + security-reviewer	Audit code, UI, tests, and security
Document	readme-updater	Update documentation with business logic

Example Flow

1. "Use the specify skill to create a PRD for team analytics."
   -> Creates a PRD document for the team analytics feature

2. "Use the plan skill to create an implementation plan for PRD-team-analytics."
   -> Appends task breakdown to the PRD

3. "Use the new-api-route skill to implement Task 1: GET /api/analytics/overview"
   -> Implements with TDD, auth, validation

4. "Use the new-feature skill to implement Task 2: Analytics dashboard page"
   -> Builds UI with React Query, loading/error states

5. "Use the code-review skill to review the analytics implementation."
   -> Architecture, performance, and pattern audit

6. "Use the security-reviewer skill to audit the analytics feature."
   -> OWASP checklist review

7. "Use the readme-updater skill to update READMEs for my recent changes."
   -> Fills documentation with business logic

---

When to Use PRDs

Use a PRD	Skip the PRD
New features spanning multiple files	Simple UI tweaks
Significant refactoring efforts	Obvious bug fixes
Complex bug investigations	Minor code changes
Changes requiring team alignment	One-file changes

Rule of thumb: If the change touches more than two or three files or requires architectural decisions, a PRD will save you time by giving the AI structured context up front.

---

Troubleshooting

Question	Answer
Should I create a PRD?	Yes if it spans multiple files or needs team alignment. No if it is a one-file change or an obvious fix.
AI not following PRD?	Enter plan mode first. Reference explicitly: “Implement PRD-X section 4.”
PRD too complex?	Start high-level and add detail later. Break into phases if needed.
Ticket integration fail	Check MCP config, verify token, ensure correct ticket ID format, or use the specify skill with ticket details.

---

Related Documentation

• Prompt Library and Skills — Agent Skills reference and effective prompts
• Build Your First Feature — See the workflow in action
• AI Development Framework — Full architecture overview