PRD Skill — Stages & Information Flow

How the PRD skill runs Socratic discovery across four depth levels, seven phases, and six reference files to produce a build-ready requirements document. Grounded in the PuRDy methodology: platform-agnostic, quantification-forward, failure-pattern-aware.

flowchart TD
    START["User asks for PRD"]
    START --> S0["Step 0: Confirm Scale
Match depth to project type"]

    S0 --> S0b["Step 0b: Deployment Target
Local / Team / Public"]

    S0b --> RC{"Reality Check
Speed vs Complexity"}

    RC -->|"Demo"| D_MIN["MINIMAL DEPTH"]
    RC -->|"MVP"| D_LIGHT["LIGHT DEPTH"]
    RC -->|"Internal Tool"| D_STD["STANDARD DEPTH"]
    RC -->|"Enterprise"| D_DEEP["DEEP DEPTH"]

    %% Minimal path
    D_MIN --> P0min["Phase 0: Skip classification"]
    P0min --> P1min["Phase 1: 2-3 questions
What, who, constraints?"]
    P1min --> P4min["Phase 4: 1 success metric"]
    P4min --> OUTmin["Minimal PRD
~5 bullet points"]

    %% Light path
    D_LIGHT --> P0lt["Phase 0: Silent classify"]
    P0lt --> P1lt["Phase 1: 3-5 min
Main pain, rough scale"]
    P1lt --> P4lt["Phase 4: 1-2 metrics
MVP scope, timeline"]
    P4lt --> P7lt["Phase 7: Generate Light PRD"]
    P7lt --> OUTlt["Light PRD
MVP template"]

    %% Standard path
    D_STD --> P0std["Phase 0: Classify & confirm
initiative-taxonomy.md"]
    P0std --> P1std["Phase 1: 7-10 min
5 Whys on primary pain
a3-framework + scale-effort"]
    P1std --> P2std["Phase 2: Stakeholder map
Approvers, maintainers, resisters
question-banks.md"]
    P2std --> P3std["Phase 3: System inventory
Integration type, PII, constraints
question-banks.md"]
    P3std --> P4std["Phase 4: 2-3 quantified metrics
Current → Target, pilot plan
scale-effort.md"]
    P4std --> P5std["Phase 5: Top 3-4 risks
Failure pattern scan
failure-patterns.md"]
    P5std --> P6std["Phase 6: Opportunity prioritization
Impact / Effort matrix"]
    P6std --> DIAG_STD{"Visual diagram?"}
    DIAG_STD -->|"Yes"| P7std["Phase 7: Generate Full PRD"]
    P7std --> OUTstd["Full PRD
prd-template.md"]

    %% Deep path
    D_DEEP --> P0dp["Phase 0: Full classify & confirm
initiative-taxonomy.md"]
    P0dp --> P1dp["Phase 1: 10-15 min
Full 5 Whys, full 3M classification
a3-framework + scale-effort"]
    P1dp --> P2dp["Phase 2: Full stakeholder map
Power scoring 1-5, RACI, missing voices
question-banks.md"]
    P2dp --> P3dp["Phase 3: Full system inventory
Data ownership, quality, transform complexity
question-banks.md"]
    P3dp --> P4dp["Phase 4: 2-3 quantified metrics
Explicit out-of-scope, pilot plan
scale-effort.md"]
    P4dp --> P5dp["Phase 5: Full 7-pattern scan
Risk + mitigation + dependency map
failure-patterns.md"]
    P5dp --> P6dp["Phase 6: Opportunity prioritization
Impact / Effort matrix"]
    P6dp --> DIAG_DP{"Visual diagram?"}
    DIAG_DP -->|"Yes"| P7dp["Phase 7: Generate Full PRD"]
    P7dp --> OUTdp["Full PRD
prd-template.md"]

    %% Shared outputs
    OUTstd --> BUILD["Offer build prompt
Copy-paste ready"]
    OUTdp --> BUILD
    BUILD --> SCAFFOLD["Repo scaffolding
git init, docs/, README"]
    SCAFFOLD --> DONE["PRD complete"]

    %% Style classes
    class D_MIN,D_LIGHT,D_STD,D_DEEP depthMinimal
    class P0min,P1min,P4min,P0lt,P1lt,P4lt,P7lt depthLight
    class P0std,P1std,P2std,P3std,P4std,P5std,P6std,P7std depthStandard
    class P0dp,P1dp,P2dp,P3dp,P4dp,P5dp,P6dp,P7dp depthDeep
    class RC,DIAG_STD,DIAG_DP decision
    class OUTmin,OUTlt,OUTstd,OUTdp,BUILD,SCAFFOLD,DONE output
    class S0,S0b phase

Minimal depth

Light depth

Standard depth

Deep depth

Decision gate

Output artifact

Phase Reference

What each phase covers, which depths include it, and which reference files inform it.

Phase 0 Context & Classification

Applies: Light • Standard • Deep initiative-taxonomy.md

5 initiative types: Data Integration, Process Automation, Tool Selection, Cross-Functional, Reporting/Analytics. Signal phrase matching, hybrid detection, gap detection by type. Standard/Deep confirm classification with user; Light notes it silently.

Phase 1 Problem Deep-Dive

Applies: All depths a3-framework.md scale-effort.md question-banks.md

Current process mapping, pain points, scale quantification, 5 Whys root cause analysis, 3M waste classification (Muda/Mura/Muri). Depth determines how deep: Minimal = 2-3 questions, Light = 3-5 min, Standard = 7-10 min with 5 Whys on primary pain, Deep = full 5 Whys on every pain point + full 3M.

Phase 2 Stakeholder Map

Applies: Standard • Deep question-banks.md

Role + decision power scoring (1-5), resistance analysis, maintainer identification, missing voices check. Standard covers key stakeholders and approvers. Deep adds full RACI, power/interest scoring, and explicit resistance strategy for every potential blocker.

Phase 3 System & Data Inventory

Applies: Standard • Deep (MVP: quick list only) question-banks.md

Each system's role, integration type (API/manual/file transfer), data sensitivity/PII flags, what can't change, integration complexity score. Deep adds data ownership mapping, quality assessment, and transformation complexity scoring.

Phase 4 Success Criteria

Applies: ALL depths (non-negotiable) scale-effort.md

Quantified metrics (current → target), MVP scope boundaries, explicit out-of-scope list. Standard/Deep add pilot plan and timeline. This is the one phase that never gets skipped — even Minimal asks "what does success look like?"

Phase 5 Risks, Constraints & Dependencies

Applies: Standard • Deep failure-patterns.md scale-effort.md

Risk + mitigation pairs, timeline/budget constraints, dependency map. 7-pattern failure scan: Integration Complexity, Ghost Town Build, Pre-Scale Paralysis, Cost Avalanche, Feature Gravity Well, Technical Debt Spiral, Reality Distortion. Standard scans top 3-4; Deep scans all 7. 40-hour threshold routes large initiatives to formal approval.

Depth Levels at a Glance

Four escalation levels controlled by project type — the skill scales depth to match scope.

Minimal

Quick Demo

Scale: Kiddie Pool, S effort

Phases: 0 skip, 1 (2-3q), 4 (1 metric)

Template: Minimal PRD

Time: 2-3 min conversation

Light

MVP

Scale: Kiddie Pool, S-M effort

Phases: 0 silent, 1 abbrev., 4, 7

Template: Light PRD

Time: 5-8 questions

Standard

Internal Tool

Scale: Olympic Pool, M-L effort

Phases: All at moderate depth

Template: Full PRD

Time: 20-30 min conversation

Deep

Platform / Enterprise

Scale: Ocean, L-XL effort

Phases: All at full depth

Template: Full PRD + visual diagram

Time: 30-45 min conversation

Reference Files

Six reference files pulled in by phase. The skill instructs reading prd-template.md before starting and scanning others during the conversation.

Reference File	Used By	Content
`prd-template.md`	Phase 7	Full and Light PRD templates with all sections
`initiative-taxonomy.md`	Phase 0	5 initiative types, signal phrases, failure patterns per type, success benchmarks
`question-banks.md`	Phases 1, 2, 3	Context-specific questions: internal tools, stakeholder, training, workflow integration
`a3-framework.md`	Phase 1	Current → Target → Gap → Root Cause → Countermeasures structure
`failure-patterns.md`	Phase 5	7 failure patterns with warning signs, prevention strategies, cascade models, immunity scorecard
`scale-effort.md`	Phases 1, 4, 5	Kiddie/Olympic/Ocean scale, S/M/L/XL effort tiers, 40-hour threshold, impact/effort matrix

Anti-Patterns the Skill Guards Against

Using Deep mode for a Kiddie Pool project — the most common failure
Jumping to solutions in Phase 1 — stay in problem space until the pain is understood
Accepting “a lot” / “better” / “faster” without numbers — quantify everything
PRD with zero open questions — you didn’t probe hard enough
Silently filling gaps in user requirements — every vague term is a question you haven’t asked
Assuming cloud for local-only projects — SQLite + localhost is the correct default
Skipping repo scaffolding — even a quick project gets git init, README, and standard structure