Complex Project Skill

When to Activate

• Task spans 3+ technical domains (frontend, backend, database, infrastructure)
• User mentions "enterprise", "production-ready", "full stack", "complete system"
• Project requires coordination between multiple specialists
• Deep codebase research would significantly improve planning
• Task is ambiguous and needs exploration before implementation
• Security, performance, or compliance requirements are mentioned
• User asks to "build from scratch" or create a "complete" solution

Workflow

Layer 0: Parallel Research Phase

Orchestrator launches up to 6 research droids in parallel:

• @codebase-researcher - Architecture patterns analysis
• @git-history-analyzer - Historical change patterns
• @library-source-reader - Technology insights
• @file-picker-agent - Targeted file discovery
• @context-researcher - Project-wide context
• @best-practices-researcher - Industry standards

Layer 1: Discovery + Intelligence

• Synthesize research findings
• Identify technical constraints and risks
• Map existing patterns and conventions

Layer 2: Planning + PRD

• Invoke @prd for comprehensive requirements
• Apply research intelligence to inform PRD
• Generate acceptance criteria with codebase context

Layer 3: Task Breakdown

• Invoke @generate-tasks for parallelized tasks
• Create independent tracks with dependency mapping
• Identify critical path and integration points (use bv --robot-insights or bv --robot-plan for graph-based validation)

Optional: Task Persistence with bd

After generating task structure, create persistent tasks using bd create:

PARENT=$(bd create "Task X.X" --description="..." -t task -p 1 --json | jq -r '.id')
bd create "Task X.Y" --description="..." -t task -p 2 --deps discovered-from:$PARENT --json

Optional: Beads Viewer Integration

After creating tasks, use bv for intelligent analysis:

# Critical path and parallel tracks
bv --robot-plan

# PageRank-based task prioritization
bv --robot-triage

# Comprehensive project health
bv --robot-insights

Layer 4: Delegation

• Orchestrator delegates to specialist droids (up to 6 parallel)
• Each droid receives full context + intelligence findings
• Tracks coordinated for maximum efficiency

Layer 5: Review & Learning

• Quality gates with intelligence validation
• Update memory files with new patterns
• Cross-project learning integration

Inputs Expected

• Project description or requirements
• Any constraints (tech stack, timeline, compliance)
• Scope boundaries (what's in/out)

Outputs Produced

• Research intelligence report
• Comprehensive PRD
• Parallelized task breakdown with tracks
• Implementation by coordinated specialist droids
• Updated memory files with learned patterns

Success Criteria

• All technical domains covered with appropriate specialists
• Parallel execution maximized (up to 6 droids)
• Research intelligence applied throughout
• Quality gates passed at each layer
• Memory updated with new patterns for future projects

When NOT to Use

• Simple single-domain tasks (use direct droid instead)
• Quick fixes or bug fixes
• Tasks with clear, simple scope
• Time-critical simple changes