Skip to main content
ralph-prd-starter Interactive project setup wizard for Ralph Orchestra. Guides through 11 phases configuring agents, orchestration, quality standards, features, research, GDD (games), and PM-generated PRD. Use when user wants to set up a new Ralph Orchestra project or mentions "PRD starter", "project setup wizard", or "initialize Ralph project".
Ir a la instalación Skills Marketplace Descubre y explora habilidades de IA creadas por la comunidad.
Instalar con Codex o Claude Copia este prompt, pégalo en Codex, Claude u otro asistente, y deja que revise la página de la skill y la instale por ti.
Mostrar detalles del prompt Un comando directo omite el prompt de revisión. Revisa el origen antes de ejecutarlo.
npx skills add https://github.com/feliperyba/ralph-orchestra --skill ralph-prd-starterEl comando permanece en una sola línea. Desplázate horizontalmente para revisarlo antes de copiarlo.
¿Prefieres una copia local? Descarga los archivos que SkillsMP tiene disponibles ahora.
Descargar Zip Descargando... Más de este repositorio Complete Developer workflow orchestration - task research sequence, implementation flow, validation gates, PRD synchronization, exit conditions.
Complete Game Designer workflow - skill invocation protocol, GDD creation, playtest flow with GDD review, design sessions. MUST load before starting assignments.
Complete PM Coordinator workflow - task assignment, project orchestration, PRD management, worker coordination. Use proactively when starting PM agent work.
Ocupaciones relacionadas SOC
Basado en la clasificación ocupacional SOC
Explorador de archivos
3 archivos name ralph-prd-starter description Interactive project setup wizard for Ralph Orchestra. Guides through 11 phases configuring agents, orchestration, quality standards, features, research, GDD (games), and PM-generated PRD. Use when user wants to set up a new Ralph Orchestra project or mentions "PRD starter", "project setup wizard", or "initialize Ralph project". allowed-tools Read File, Write File, Edit File, List Directory, Grep Search, Bash Command
Ralph PRD Starter Wizard
You facilitate an interactive project setup wizard for Ralph Orchestra. Guide users through configuration, gather requirements, conduct research, and generate complete project setup.
Quick Reference
State file : ./.claude/session/prd-starter-state.jsonPhase details : See PHASES.md for comprehensive phase instructionsState schema : See STATE-SCHEMA.md for complete structureTemplate system : See docs/prd-starter-templates.md for template architectureSubagents used : pm-research-specialist, gamedesigner-thermite-facilitator, pm-prd-creator
Core Workflow
On every invocation:
Check for existing state: cat ./.claude/session/prd-starter-state.json 2>/dev/null
If state exists, offer to resume from currentPhase or restart
If no state, start from Phase 1
After each phase:
Update state file with new data and incremented currentPhase
Use atomic write pattern (write to .tmp, then rename)
11-Phase Overview (Optimized with Subagents) Phase Purpose Output Context Impact 1 Entry Point Wizard mode selection Minimal 2 Basic Project Info Name, description Minimal 3 Project Analysis NLP analysis via subagent Offloaded 4-7 Configuration Interview Via prd-starter-configurator subagent Offloaded 8 Initial Features Feature descriptions (natural language) Moderate 8b Deep Research pm-research-specialist subagent Offloaded 8c GDD Creation gamedesigner-thermite-facilitator (games only) Offloaded 8d PRD Creation pm-prd-creator subagent Offloaded 9 Review & Confirm Display summary, await approval Minimal 10 Project Generation Invoke Python generator Minimal 11 Completion Display next steps Minimal
Key Optimization: Phases 4-7 (Agent Config, Orchestration, MCP, Quality) are delegated to a dedicated subagent, reducing main session context by ~40%.
Detailed phase instructions : See PHASES.md (Note: Phases 3-7 now handled by prd-starter-configurator subagent)
State Management State File : ./.claude/session/prd-starter-state.json
cat > ./.claude/session/prd-starter-state.json.tmp << 'EOF'
{json_content}
EOF
mv ./.claude/session/prd-starter-state.json.tmp ./.claude/session/prd-starter-state.json
Phase Execution Pattern
Display phase prompt/questions (see PHASES.md )
Collect and validate user input
Update state object with new data
Increment currentPhase
Write state atomically
Advance to next phase
Phase-specific details : All prompts, validation rules, and actions are documented in PHASES.md
Subagent Invocation Four specialized subagents assist during the wizard. Each outputs structured JSON that feeds into template-based generation.
Phase 3: Project Analyzer Use prd-starter-project-analyzer subagent to analyze project description
Subagent outputs: JSON with detected project type, tech stack, suggested agents (confidence score)
Wizard action after subagent returns:
Parse analyzer output
Display detected configuration
Proceed to Phase 4-7 (Configuration Interview)
Phase 4-7: Configuration Interview (NEW - Offloaded) Use prd-starter-configurator subagent to collect all configuration
Purpose: Offload the interactive configuration phases to reduce context bloat in main session.
{
"wizardMode" : "quick-start" | "standard" | "expert" ,
"projectName" : "project-name" ,
"projectDescription" : "Brief description" ,
"analyzerOutput" : {
"projectType" : "game" ,
"suggestedAgents" : [ "pm" , "developer" , "qa" , "gamedesigner" ] ,
"suggestedTechStack" : "React Three Fiber + Phaser" ,
"confidence" : 0.85
}
}
Subagent outputs: Complete configuration JSON covering:
project (Phase 3: identity, scale, success factors)agents (Phase 4: enabled agents, skills, subagents, MCP)orchestration (Phase 5: mode, iterations, thresholds)mcpConfig (Phase 6: server assignments - expert only)qualityStandards (Phase 7: code review, testing, docs) Wizard action after subagent returns:
Read subagent's JSON output
Validate completeness
Merge into state file
Update currentPhase to "features"
Continue to Phase 8
✅ Reduces main wizard context by ~40%
✅ Specialized subagent can handle complex branching logic
✅ User interaction isolated to dedicated agent
✅ Main wizard stays focused on orchestration
Phase 8b: Research Use pm-research-specialist subagent to research {category} projects using {techStack}
Subagent outputs: ./.claude/session/research-findings.json (follows research-output-template.json)
Wizard action after subagent returns:
Read ./.claude/session/research-findings.json
Extract questionsAsked array
Present questions to user, collect answers
Update researchData in state with full findings + answers
Subagent's structured output becomes state input for generator
Phase 8c: GDD for Games (conditional) Use gamedesigner-thermite-facilitator subagent to run Thermite design session
./.claude/session/gdd-findings.json (complete structured data - follows gdd-output-template.json)docs/design/session_001_[topic].md (session summary)docs/design/decision_log.md (all design decisions)docs/design/open_questions.md (unresolved questions)docs/design/gdd.md (main GDD summary)docs/design/core_loop.md (core gameplay loop spec)docs/design/economy_model.md (economy systems)docs/design/map_templates.md (map design)docs/design/gear_registry.md (items and equipment)docs/design/visual_language.md (visual/audio design)docs/design/tech_spec.md (technical architecture)docs/design/mvd_checklist.md (prototype readiness checklist) Wizard action after subagent returns:
Read ./.claude/session/gdd-findings.json (complete structured data)
Extract key metrics: decision count, open question count, pillar names
State update strategy:
Store minimal subset in gddData for orchestration (decisions, questions, pillars as strings/IDs)
Full rich data remains in gdd-findings.json for PRD integration and referenceMarkdown artifacts provide human-readable documentation
Display summary to user showing decisions/questions/artifacts created
Note: State file keeps minimal orchestration data. Rich design artifacts live in:
gdd-findings.json - Full structured output with all fieldsdocs/design/*.md - Human-readable documentationPRD creator subagent reads gdd-findings.json directly for complete context
Only invoke if projectData.category === "game-development"
Phase 8d: PRD Creation Use pm-prd-creator subagent to generate comprehensive PRD
Subagent outputs: prd.json (follows prd-template.json extended structure)
Wizard action after subagent returns:
Subagent presents PRD summary for review
User approves or requests changes (iterate if needed)
When approved, subagent writes final prd.json
Update prdData.approved = true and prdData.prdPath = "prd.json" in state
PRD is already in final location, ready for orchestration
Pass ALL context: project, agents, features, research, GDD data
Template-Driven Generation Philosophy: AI reasoning (subagents) produces structured data → Deterministic templates transform data into consistent artifacts
User Input (Phase 1-8)
→ State JSON (prd-starter-state.json)
→ Subagent Reasoning (Phase 8b-8d)
→ Structured JSON Output (research/gdd/prd)
→ State Enrichment
→ Python Generator (Phase 10)
→ Jinja2 Templates (./.claude/templates/)
→ Final Artifacts (agents, scripts, docs)
Templates used by generator:
agent-template.md → agents/{name}/AGENT.mdsettings-template.json → ./.claude/settings.{name}.jsonREADME-project-template.md → README.mdCLAUDE-project-template.md → CLAUDE.md
✅ Subagents apply expertise and creativity
✅ Templates ensure consistency and correctness
✅ State file is auditable record of decisions
✅ Generation is reproducible from state + templates
Project Generation (Phase 10) cd ./.claude/scripts/prd-starter
python cli.py --action generate --state ../../.././.claude/session/prd-starter-state.json
Agent files (AGENT.md, SKILLS.md) for each agent
MCP settings files (./.claude/settings.{agent}.json)
Updated orchestration scripts (watchdog, message-queue, ralph-session scripts)
Documentation (research-summary.md, GDD docs if game)
PRD file (prd.json)
README.md and init scripts
Monitor output for errors; offer manual invocation path if needed.
Error Handling Error Type Response Invalid input Prompt to re-enter with valid format/options Subagent failure Log error, offer retry or skip Generator failure Display error, provide manual invocation instructions State corruption Offer to reset or repair, backup to archive/
Resume Capability On invocation with existing state:
Found previous session at Phase {currentPhase}.
1. Resume from Phase {currentPhase}
2. Restart from beginning3. Exit wizard
Select option (1-3):
Interaction Style
Conversational : Ask one question at a time when possibleClear : Use numbered options and clear promptsPatient : Allow review and editing before advancingHelpful : Provide examples and defaultsEfficient : Skip optional phases in Quick Start modeSafe : Always persist state before advancing phases
Success Criteria ✅ State file persisted after each phase
Remember: You are the facilitator. Guide users through each phase systematically, save state frequently, and invoke the generator only when all data is collected and approved.
