Claude Skill

ue5-module-router

Route UE5.6/UE5.7 questions to the most precise skill using module names, aliases, intent keywords, and layer context. Works for explicit module prompts (RenderCore, AIModule, AssetRegistry) and natural language requests.

LLM Mart · 0 points · 7 0 0

Virus-scanned Reviewed automatically before listing.

Full trust report

Download teixasalone-unrealengine5-skills-skills_ue5-module-router-397623a.zip · 33 KB

teixasalone/unrealengine5-skills 17 3 forks MIT Updated 7d ago

Part of teixasalone/unrealengine5-skills — 3 skills

Install

skills CLI npx skills add https://github.com/teixasalone/UnrealEngine5-Skills/tree/main/skills/ue5-module-router

Claude Code

claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install teixasalone-unrealengine5-skills@llmmart

Git git clone https://github.com/teixasalone/UnrealEngine5-Skills.git

The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole teixasalone/unrealengine5-skills collection as a plugin from our marketplace. Git is the plain clone.

Skill manifest

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If mapping is outdated, regenerate from ue5-architecture/scripts/generate_module_index_v2.py.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.

Files (unrealengine5-skills)

agents

openai.yaml 302 B

display_name: UE5.6/UE5.7 Module Router
short_description: Route UE5.6/UE5.7 requests to the right skill by exact module name and alias mapping.
default_prompt: Identify the UE5.6/UE5.7 module(s) in this request, route to the most precise skill using the routing table, and explain routing confidence.

references

routing-policy.md 1.9 KB

# Routing Policy (UE5.6/UE5.7)

## Priority Order
1. Exact module name match in routing table.
2. Exact Build.cs path segment match.
3. Alias/keyword match from routing table.
4. Domain fallback from module index v2.

## Confidence Levels
- High: `route_reason=module_override` or exact module hit with stable mapping.
- Medium: `route_reason=domain_mapping` with non-`General` domain.
- Low: fallback on `General` domain + weak alias evidence.

## Output Contract
- `primary_skill`
- `secondary_skill` (if present)
- `recommended_mcp_tools[]`
- `matched_modules[]`
- `route_reason`
- `route_confidence`

## Tool Priority Matrix
- Blueprint requests:
  - Prefer: `blueprint_feature_build`, `blueprint_modify`, `blueprint_query`
  - Then: `enhanced_input`, `get_output_log`
  - Last resort: `execute_script`
- World interaction requests:
  - Prefer: `spawn_actor`, `get_level_actors`, `set_property`, `move_actor`, `delete_actors`, `open_level`
  - Then: `capture_viewport`, `get_output_log`
  - Last resort: `execute_script`
- Save/load/network requests:
  - Prefer: `character_data`, `blueprint_query`, `blueprint_modify`, `asset`, `asset_search`
  - Then: `get_output_log`, `task_*`
  - Last resort: `execute_script`
- Performance/packaging requests:
  - Prefer: `run_console_command`, `get_output_log`, `capture_viewport`, `open_level`, `asset_dependencies`, `asset_referencers`
  - Then: `task_*`
  - Last resort: `execute_script`
- Debug/validation requests:
  - Prefer: `get_output_log`, `asset_search`, `blueprint_query`, `get_level_actors`, `task_list`, `task_status`, `task_result`
  - Then: `capture_viewport`
  - Last resort: `execute_script`

## execute_script Fallback Conditions
Use `execute_script` only when all conditions are met:
1. No dedicated tool supports the required operation.
2. The request cannot be decomposed into existing MCP tool calls.
3. The reason is explicitly stated in the response.
4. Script scope is minimal and reversible where possible.

ue5-module-routing-table-draft.deprecated.csv 109.1 KB · in bundle
ue5-module-routing-table-final.csv 136.3 KB · in bundle

SKILL.md 1.8 KB

---
name: ue5-module-router
description: Route UE5.6/UE5.7 questions to the most precise skill using module names, aliases, intent keywords, and layer context. Works for explicit module prompts (RenderCore, AIModule, AssetRegistry) and natural language requests.
---

# 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.