Quick Start

• Extract explicit module names from user prompt first.
• If module is found, route by exact module mapping before keyword heuristics.
• If module is not found, use aliases and layer context.

Workflow

• Parse prompt for module candidates (for example RenderCore, AIModule, AssetRegistry).
• Lookup module in ue5-module-routing-table-final.csv.
• If multiple hits, prioritize:
  1. exact module name match
  2. Build.cs path similarity
  3. alias overlap with prompt keywords
• Return routing payload:
  • primary_skill
  • secondary_skill
  • recommended_mcp_tools[]
  • route_confidence
  • route_reason
• Use secondary_skill when request spans multiple concerns in one module context.

Constraints

• Prefer deterministic routing; avoid broad guesses if exact module match exists.
• Keep one primary target skill unless user explicitly asks cross-module analysis.
• If module maps to ue5-architecture, answer module-boundary/design first.
• Prefer dedicated MCP tools before execute_script.

Failure Handling

• If no module is recognized, fallback to closest capability skill and state reason.
• If confidence is low, provide top 2 candidates and request module confirmation.
• Use execute_script only when dedicated MCP tools are insufficient and explain why.
• If mapping is outdated, regenerate from ue5-architecture/scripts/generate_module_index_v2.py.

Escalation

• Escalate for large cross-cutting refactors across many modules.
• Escalate when requested module belongs to plugin source outside indexed scope.