/moe-agents - Mixture of Experts Assembly

Analyzes your use case or workflow and assembles an optimal team of specialized CODITECT agents, each contributing their unique expertise.

System Prompt

⚠️ EXECUTION DIRECTIVE: When the user invokes this command, you MUST:

IMMEDIATELY execute - no questions, no explanations first
ALWAYS show full output from script/tool execution
ALWAYS provide summary after execution completes

DO NOT:

Say "I don't need to take action" - you ALWAYS execute when invoked
Ask for confirmation unless requires_confirmation: true in frontmatter
Skip execution even if it seems redundant - run it anyway

The user invoking the command IS the confirmation.

Usage

/moe-agents <use-case-description>

How It Works

When you run /moe-agents, I will:

Load Project Context (ADR-156) - Detect current project and load relevant context
Analyze the Use Case - Break down the task into component requirements
Identify Required Expertise - Map requirements to agent specializations
Assemble Expert Team - Select 3-7 agents with complementary skills
Define Roles & Responsibilities - Assign specific tasks to each expert
Inject Project Context - Pass relevant decisions and patterns to agents
Create Coordination Plan - Orchestrate how experts will collaborate
Generate Execution Commands - Ready-to-run /agent invocations with project context

Project Context Loading

Before assembling experts, the command loads project-specific context:

# Step 1: Detect project
cd /Users/halcasteel/PROJECTS/coditect-rollout-master/submodules/core/coditect-core
source .venv/bin/activate
python3 -c "from scripts.core.paths import discover_project; print(discover_project() or 'global')"

# Step 2: Load project context
python3 scripts/context-query.py --project-stats
python3 scripts/context-query.py --decisions --limit 10 --filter-project
python3 scripts/context-query.py --errors --limit 5 --filter-project

Context passed to agents:

Project ID - Ensures agents operate in correct scope
Recent Decisions - Architectural choices to respect
Error Patterns - Known issues and solutions
Skill Learnings - Accumulated best practices

Examples

Example 1: Build a SaaS Product

/moe-agents build a complete SaaS product from idea to launch

Expert Team Assembled:

Role	Agent	Responsibility
Lead Architect	`senior-architect`	System design, tech stack decisions
Business Strategist	`business-intelligence-analyst`	Market analysis, pricing strategy
Backend Expert	`backend-architect`	API design, database schema
Frontend Expert	`frontend-react-typescript-expert`	UI/UX implementation
DevOps Lead	`devops-engineer`	CI/CD, infrastructure
Security Expert	`security-specialist`	Security audit, compliance
Documentation	`codi-documentation-writer`	User docs, API docs

Coordination:

/agent orchestrator "Coordinate SaaS development with senior-architect leading architecture, backend-architect on APIs, frontend-react-typescript-expert on UI, devops-engineer on infrastructure, and security-specialist on security review"

Example 2: Competitive Market Analysis

/moe-agents comprehensive competitive analysis for AI code assistant market

Expert Team Assembled:

Role	Agent	Responsibility
Lead Analyst	`competitive-market-analyst`	Competitive landscape, positioning
Market Researcher	`market-researcher`	Market size, trends, growth
Business Analyst	`business-intelligence-analyst`	Financial analysis, unit economics
Web Researcher	`web-search-researcher`	Current pricing, features, news
Strategy Writer	`strategy-brief-generator`	Strategic recommendations

Example 3: Security Audit

/moe-agents complete security audit for production API

Expert Team Assembled:

Role	Agent	Responsibility
Security Lead	`security-specialist`	Overall security assessment
Penetration Tester	`penetration-testing-agent`	Vulnerability testing
API Security	`backend-api-security`	API-specific vulnerabilities
Code Reviewer	`code-reviewer`	Secure coding practices
Compliance	`compliance-checker-agent`	Regulatory compliance

Example 4: Documentation Overhaul

/moe-agents complete documentation overhaul for developer platform

Expert Team Assembled:

Role	Agent	Responsibility
Doc Lead	`codi-documentation-writer`	Overall documentation strategy
API Docs	`code-documentation`	API reference documentation
Architecture	`software-design-architect`	Architecture diagrams, C4 models
Quality Review	`documentation-quality-agent`	Quality, consistency review
Librarian	`documentation-librarian`	Organization, cross-references

Expert Selection Matrix

By Domain

Domain	Primary Experts	Supporting Experts
Business	`business-intelligence-analyst`, `competitive-market-analyst`	`market-researcher`, `venture-capital-business-analyst`
Architecture	`senior-architect`, `software-design-architect`	`backend-architect`, `database-architect`
Development	`code-reviewer`, `debugger`	Domain-specific (React, Rust, etc.)
Security	`security-specialist`, `penetration-testing-agent`	`backend-api-security`, `compliance-checker-agent`
DevOps	`devops-engineer`, `cloud-architect`	`k8s-statefulset-specialist`, `monitoring-specialist`
Documentation	`codi-documentation-writer`, `documentation-generation`	`documentation-quality-agent`
Research	`research-agent`, `web-search-researcher`	`market-researcher`, `competitive-analyst`

By Task Complexity

Complexity	Team Size	Coordination
Simple	1-2 experts	Direct invocation
Moderate	3-4 experts	Sequential workflow
Complex	5-7 experts	Orchestrator coordination
Enterprise	7+ experts	Multi-phase orchestration

Output Format

When you run /moe-agents, you'll receive:

Project Context Header - Current project and loaded context (ADR-156)
Team Roster - Selected experts with roles
Task Breakdown - What each expert will handle
Execution Plan - Recommended order of operations
Agent Commands - Ready-to-run /agent invocations with project context
Orchestration Command - Single command to coordinate all experts

Output Structure

┌─────────────────────────────────────────────────────────────┐
│  /moe-agents: <use-case>                                    │
│  Project: <project-id or "global">  (ADR-156)               │
├─────────────────────────────────────────────────────────────┤
│  PROJECT CONTEXT LOADED                                      │
│  ─────────────────────                                       │
│  Decisions: N loaded (influences architecture choices)       │
│  Error Patterns: N loaded (known issues to avoid)            │
│  Learnings: N loaded (best practices applied)                │
│                                                              │
├─────────────────────────────────────────────────────────────┤
│  EXPERT TEAM ASSEMBLED                                       │
│  ─────────────────────                                       │
│  Lead: <agent-name> - <responsibility>                      │
│  Expert 2: <agent-name> - <responsibility>                  │
│  Expert 3: <agent-name> - <responsibility>                  │
│  ...                                                         │
├─────────────────────────────────────────────────────────────┤
│  EXECUTION COMMANDS                                          │
│  ──────────────────                                          │
│  /agent <name> "<task>" --project <project-id>              │
│  ...                                                         │
├─────────────────────────────────────────────────────────────┤
│  ORCHESTRATION                                               │
│  /agent orchestrator "coordinate: <plan> [project: X]"      │
└─────────────────────────────────────────────────────────────┘

Combining with /moe-judges

For critical tasks, combine expert assembly with judge evaluation:

# Step 1: Assemble experts
/moe-agents build enterprise API

# Step 2: Execute with experts
/agent orchestrator "coordinate API development..."

# Step 3: Judge the output
/moe-judges evaluate the API design for production readiness

Or use /moe-workflow for automatic expert + judge combination.

/which <task> - Quick single-agent recommendation
/moe-judges <output> - Evaluate output with judge panel
/moe-workflow <task> - Full expert + judge workflow
/agent <name> <task> - Invoke specific agent

Action Policy

<default_behavior> This command analyzes and assembles without executing. Provides:

Expert team roster
Role assignments
Coordination plan
Ready-to-run commands

User decides when to execute the assembled team. </default_behavior>

After assembly, verify: - Team size appropriate (3-7) - Roles non-overlapping - Commands generated - Coordination plan clear

Success Output

When moe-agents completes:

✅ COMMAND COMPLETE: /moe-agents
Project: <project-id or "global">
Use Case: <description>
Team Size: N experts
Lead: <primary-agent>
Context: N decisions, N patterns loaded
Commands: N generated (with project context)
Coordination: Ready

Completion Checklist

Before marking complete:

Failure Indicators

This command has FAILED if:

❌ No use case provided
❌ No experts selected
❌ Commands not generated
❌ Role conflicts exist

When NOT to Use

Do NOT use when:

Single agent sufficient (use /which)
Need immediate execution
Simple task (use /agent directly)

Anti-Patterns (Avoid)

Anti-Pattern	Problem	Solution
Too many experts	Coordination overhead	Limit to 5-7 for complex
Overlapping roles	Confusion	Clear responsibility matrix
Skip coordination	Chaos	Use orchestrator for >3

Principles

This command embodies:

#6 Clear, Understandable - Clear role matrix
#3 Complete Execution - Full team assembly
#9 Based on Facts - Evidence-based selection

Full Standard: CODITECT-STANDARD-AUTOMATION.md

Project Context Integration (ADR-156)

The /moe-agents command now loads and applies project-specific context:

Context Sources

Source	Database	Purpose
Project ID	`discover_project()`	Scope all operations
Decisions	`org.db:decisions`	Architectural constraints
Error Patterns	`org.db:error_solutions`	Known issues to avoid
Skill Learnings	`org.db:skill_learnings`	Best practices

How Context Influences Assembly

Agent Selection - Prefer agents that have worked on this project
Role Assignment - Consider past decisions when assigning responsibilities
Coordination Plan - Incorporate lessons from error patterns
Command Generation - Include project context in /agent invocations

Context-Aware Commands

Generated commands include project context:

# Without project context (old)
/agent senior-architect "design API for user service"

# With project context (ADR-156)
/agent senior-architect "design API for user service" \
  --context "Project: PILOT, Decisions: Use Django REST, PostgreSQL"

Usage

/moe-agents build authentication system    # Uses auto-detected project
CODITECT_PROJECT=my-proj /moe-agents ...   # Explicit project

Version: 1.1.0 Created: 2025-12-22 Updated: 2026-02-04 Author: CODITECT Team ADR: ADR-156

System Prompt​

Usage​

How It Works​

Project Context Loading​

Examples​

Example 1: Build a SaaS Product​

Example 2: Competitive Market Analysis​

Example 3: Security Audit​

Example 4: Documentation Overhaul​

Expert Selection Matrix​

By Domain​

By Task Complexity​

Output Format​

Output Structure​

Combining with /moe-judges​

Related Commands​

Action Policy​

Success Output​

Completion Checklist​

Failure Indicators​

When NOT to Use​

Anti-Patterns (Avoid)​

Principles​

Project Context Integration (ADR-156)​

Context Sources​

How Context Influences Assembly​

Context-Aware Commands​

Usage​

System Prompt

Usage

How It Works

Project Context Loading

Examples

Example 1: Build a SaaS Product

Example 2: Competitive Market Analysis

Example 3: Security Audit

Example 4: Documentation Overhaul

Expert Selection Matrix

By Domain

By Task Complexity

Output Format

Output Structure

Combining with /moe-judges

Related Commands

Action Policy

Success Output

Completion Checklist

Failure Indicators

When NOT to Use

Anti-Patterns (Avoid)

Principles

Project Context Integration (ADR-156)

Context Sources

How Context Influences Assembly

Context-Aware Commands

Usage