KN Standard PRD

Purpose

Produce execution-ready PRDs that align business, product, design, and engineering. This skill combines strict requirement quality, practical templates, and deterministic validation.

When To Use

Use when the user asks to:

• Create or improve a PRD
• Document feature requirements
• Define scope and acceptance criteria
• Produce lean, one-pager, technical, or AI-feature PRDs

Operating Principles

• Ask before drafting: do not write a full PRD without discovery.
• Be measurable: replace vague words with thresholds and target values.
• Separate what/why from how: avoid over-specifying implementation.
• Guard scope: explicitly list in-scope and out-of-scope.
• Preserve uncertainty: unknowns are written as TBD with owner and due date.

Workflow

1) Discovery Interview (Required)

Ask focused questions and fill missing context. Use references/discovery_question_bank.md.

Minimum discovery bar:

• If user context is vague: ask at least 5 questions.
• If user gives a detailed brief: ask at least 2 critical gap questions.

Required topics:

• Problem and urgency
• Target users and use cases
• Business outcomes and success metrics
• Constraints (timeline, legal, platform, budget)
• Scope boundaries and non-goals

2) Select PRD Mode

Choose one mode with the user:

• standard: end-to-end production PRD
• lean: agile-friendly compact PRD
• one-pager: executive-level brief
• technical: engineering-heavy requirements
• ai-feature: adds model/tool/evaluation sections

Template sources:

• references/prd_template_standard.md
• references/prd_template_lean.md

3) Draft Requirements

For each major capability:

• Write user stories in standard format.
• Add acceptance criteria that are testable.
• Add non-functional requirements (performance, reliability, security, accessibility).

Story patterns: references/user_story_examples.md

4) Define Metrics

Pick a framework using references/metrics_framework_selector.md.

• Always define one North Star metric.
• Include leading and lagging indicators.
• Add baseline, target, and timeframe.

5) Quality Gate (Required)

Run validation:

scripts/validate_prd.sh <file.md> --mode <standard|lean|one-pager|technical|ai-feature>

Compatibility note:

• scripts/validate_prd.sh is a Bash script and must be run with Bash (bash scripts/validate_prd.sh ...).
• If Bash is unavailable in the environment, skip script execution and run a manual checklist from references/prd_review_checklist.md.

If validation reports critical issues, revise before presenting as final.

Output Contract

A final PRD must contain at minimum:

• Problem statement
• Goals/objectives
• User stories with acceptance criteria
• Success metrics (with numeric targets)
• Scope (in + out)
• Risks and dependencies

For ai-feature mode, also include:

• Model/provider assumptions
• Evaluation dataset/benchmark plan
• Safety/privacy constraints
• Fallback behavior on low confidence

Requirement Quality Standard

Bad:

• "Search should be fast"

Good:

• "p95 search latency <= 250ms on 100k records"

Bad:

• "Onboarding should be intuitive"

Good:

• "Activation rate >= 45% within first session"

Anti-Patterns To Reject

• Writing PRD without discovery
• Missing out-of-scope section
• Acceptance criteria that are not testable
• Metrics without baseline/target/time window
• Placeholder-heavy drafts presented as final

Bundled Resources

references/

• prd_template_standard.md: full PRD structure
• prd_template_lean.md: compact structure for smaller changes
• discovery_question_bank.md: interview questions by scenario
• metrics_framework_selector.md: framework selection and examples
• mode_selection_matrix.md: how to choose the right PRD mode quickly
• user_story_examples.md: story and acceptance-criteria patterns
• prd_review_checklist.md: pre-signoff quality checklist

scripts/

• validate_prd.sh: structure and quality checks with mode support

Execution Notes

• Prefer concise, skimmable markdown.
• Keep stakeholder language clear and implementation-neutral.
• Where confidence is low, explicitly mark assumptions and open questions.