Skip to main content

Command Quick Reference

192 commands at your fingertips. This guide gets you productive in 10 minutes.


Your First 5 Commands

Start every CODITECT session with these:

#CommandWhat It DoesWhen to Use
1/orientSession orientation - loads context, finds plansStart of every session
2/which <task>Find the best agent for your taskBefore any task
3/agent <name> "task"Invoke any of 157 agentsExecute tasks
4/commitCreate git commits with smart messagesAfter making changes
5/cxSave session to memory databaseEnd of every session
# Perfect session workflow
/orient # Start
/which build REST API # Find agent
/agent senior-architect "design REST API for user management"
/commit # Save changes
/cx # End - save context

The 5W Discovery Framework

CODITECT provides 6 discovery commands (5W + HOW) for finding anything:

/what - "What exists?"

/what                           # Component inventory (2,172 total)
/what agents # List all 157 agents
/what commands # List all 192 commands
/what skills # List all 223 skills
/what can orchestrator # What can this agent do?
/what is /classify # Full details about a command
/what --search "security" # Search all components

/which - "Which agent for this task?"

/which deploy to kubernetes     # Find best agent → k8s-specialist
/which review security # Find best agent → security-specialist
/which create API docs # Find best agent → codi-documentation-writer
/which build business plan # Find best agent → business-analyst
/which coordinate complex task # Find best agent → orchestrator

Output includes:

  • Primary recommendation with match score
  • Supporting agents with alternatives
  • Ready-to-use /agent invocation command
  • "When to use which" guidance

/how - "How do I do this?"

/how use orchestrator           # Step-by-step agent usage
/how create new project # Project creation guide
/how setup CI/CD # CI/CD pipeline guide
/how write tests # Testing guidance

/why - "Why was this decided?"

/why use SQLite for context     # References ADR explaining decision
/why track nomenclature # Architecture decision rationale

/where - "Where is this located?"

/where orchestrator             # File path: agents/orchestrator.md
/where /classify # Script: scripts/moe_classifier/

/when - "When should I use this?"

/when use plan mode             # Best scenarios for plan mode
/when use orchestrator # When multi-agent coordination is needed

Agent Discovery with /which

The /which command is your gateway to 157 specialized agents. Master these patterns:

Basic Usage

/which <natural language description of your task>

Common Searches

Task TypeCommandRecommended Agent
Code review/which review codecode-reviewer
Security audit/which security vulnerabilitysecurity-specialist
API development/which build REST APIsenior-architect
Documentation/which write docscodi-documentation-writer
Testing/which create teststesting-specialist
DevOps/which deploy kubernetesk8s-statefulset-specialist
Database/which design databasedatabase-architect
Business/which business planbusiness-analyst
Research/which market researchmarket-researcher
Multi-task/which coordinate multiple tasksorchestrator

Advanced /which Patterns

# Find orchestrators for complex tasks
/which moe-agents orchestrator project-organizer

# Find multiple agents for a workflow
/which analyze pilot tasklist

# Combine with domain
/which security specialist for healthcare

# Find by capability
/which agent that can coordinate other agents

Understanding /which Output

┌─────────────────────────────────────────────────────────────┐
│ /which: deploy to kubernetes │
├─────────────────────────────────────────────────────────────┤
│ PRIMARY RECOMMENDATION │
│ ────────────────────── │
│ Agent: k8s-statefulset-specialist │
│ Match: 94% │
│ Health: [85%] ↑ │
│ Why: Specialized in Kubernetes deployments │
│ │
│ Invocation: │
│ /agent k8s-statefulset-specialist "deploy application" │
├─────────────────────────────────────────────────────────────┤
│ WHEN TO USE WHICH │
│ • Use k8s-specialist when: K8s-specific work │
│ • Use devops-engineer instead when: Full CI/CD pipeline │
│ • Use cloud-architect instead when: Provision cluster │
└─────────────────────────────────────────────────────────────┘

The MoE System (Mixture of Experts)

CODITECT's MoE system provides multi-expert evaluation and execution:

Core MoE Commands

CommandPurposeWhen to Use
/moe-agents <task>Assemble expert team to CREATEBuilding something
/moe-judges <target>Assemble judge panel to EVALUATEReviewing quality
/moe-workflow <task>Combined create + evaluateFull quality cycle

/moe-agents - Build with Experts

# Assemble expert team for implementation
/moe-agents build authentication system
/moe-agents create API documentation
/moe-agents design database schema

What happens:

  1. Analyzes task requirements
  2. Selects 3-5 expert agents
  3. Each expert contributes from their specialty
  4. Orchestrator coordinates output

/moe-judges - Evaluate with Multi-Model Panel

# Evaluate deliverables with diverse perspectives
/moe-judges evaluate API design for production readiness
/moe-judges evaluate business plan for investor readiness
/moe-judges evaluate security implementation

Judge panel includes:

Judge TypeModels UsedEvaluates
Technical ArchitectClaude SonnetDesign patterns, architecture
Security AuditorClaude Opus, GPT-4oVulnerabilities, OWASP
Compliance AuditorGPT-4oRegulatory compliance
QA EvaluatorClaude HaikuTest coverage, quality
Domain ExpertDeepSeek, QwenDomain-specific concerns

Output includes:

  • Weighted scores (1-10) per dimension
  • Overall verdict (APPROVED, CONDITIONAL, REVISION REQUIRED)
  • Required actions for improvement

/moe-workflow - Complete Cycle

# Create AND evaluate in one workflow
/moe-workflow build and evaluate authentication system

Executes:

  1. /moe-agents to build
  2. /moe-judges to evaluate
  3. Iteration if below threshold

Other MoE Commands

/moe-calibrate                  # Calibrate judge panel
/moe-learn # Learn from judge feedback
/moe-analyze <target> # Multi-model analysis
/classify <path> # Document classification with MoE

Git Workflow Commands

CODITECT streamlines all git operations:

Essential Git Commands

CommandWhat It Does
/commitSmart commit with conventional format
/git-syncSync with remote, handle conflicts
/describe-prGenerate PR title and description
/pr-enhanceEnhance existing PR
/smart-mergeIntelligent merge conflict resolution

/commit - Smart Commits

/commit                         # Auto-analyze and propose commits
/commit --co-author # Add Claude co-author attribution
/commit "Add user authentication" # With specific message

What happens:

  1. Analyzes git status and diff
  2. Groups related changes logically
  3. Generates conventional commit messages
  4. Presents plan for approval
  5. Creates commits

/git-sync - Complete Sync

/git-sync                       # Sync current branch
/git-sync --target all # Sync all submodules
/git-sync --target main # Sync to main branch

/describe-pr - PR Generation

/describe-pr                    # Generate PR title and description
/describe-pr 123 # For specific PR number

Output:

## Summary
- Added user authentication with JWT
- Implemented role-based access control
- Added comprehensive test suite

## Test Plan
- [ ] Run unit tests: `npm test`
- [ ] Verify JWT token generation
- [ ] Test role permissions

Git Workflow Pattern

# Complete git workflow
/orient # Start session
# ... make changes ...
/commit # Create commits
/describe-pr # Generate PR
# Push and create PR

Command Chaining Patterns

String commands together for powerful workflows:

Session Workflow (Daily)

# Morning start
/orient

# Find and execute task
/which <your task>
/agent <recommended-agent> "<task details>"

# Review your work
/moe-judges evaluate <what you built>

# Save and commit
/commit
/cx

Project Creation Workflow

/project-new "REST API Service"    # Create project
/which design API # Find architect
/agent senior-architect "design REST API for user management"
/agent security-specialist "review API design for vulnerabilities"
/commit

Quality Assurance Workflow

# Build
/moe-agents build authentication system

# Evaluate
/moe-judges evaluate authentication for production

# If issues found, iterate
/agent security-specialist "fix rate limiting"
/moe-judges evaluate authentication for production

Documentation Workflow

/which write API docs
/agent codi-documentation-writer "create API documentation"
/moe-judges evaluate documentation for completeness
/commit

Research-then-Implement Pattern

/research "best practices for JWT authentication"
/which implement authentication
/agent senior-architect "implement JWT auth based on research"

Context & Memory Commands

Combat catastrophic forgetting with these commands:

/cx - Save Context (Session End)

/cx                             # Process all sessions into memory
/cx --with-embeddings # Include semantic embeddings
/cx --dry-run # Preview what would be saved

/cxq - Query Memory (Session Start/During)

# Search
/cxq "authentication" # Full-text search
/cxq --semantic "user login" # Semantic search

# Knowledge retrieval
/cxq --decisions # Past decisions made
/cxq --patterns --language python # Code patterns
/cxq --errors "TypeError" # Error-solution pairs

# Recent activity
/cxq --recent 20 # Last 20 messages
/cxq --today # Today's messages
/cxq --sessions # List all sessions

# Statistics
/cxq --stats # Database statistics

Memory Workflow

# Start of session - recall context
/orient
/cxq --recall "current project topic"
/cxq --decisions --recent 10

# During work - search past solutions
/cxq --errors "similar error message"
/cxq --patterns "authentication"

# End of session - save everything
/cx

Planning & Analysis Commands

Planning Commands

CommandPurpose
/deliberationPure planning, no code
/strategyArchitectural planning with C4 diagrams
/create-planCreate implementation plan
/validate-planValidate plan before execution
/implement-planExecute plan step-by-step

Analysis Commands

CommandPurpose
/analyzeCode review with rubrics
/researchVerify assumptions, check APIs
/research-codebaseSearch codebase patterns

Planning Pattern

# Research first
/research "best approach for feature X"

# Plan
/deliberation "plan feature X implementation"
# or
/create-plan "feature X"

# Validate
/validate-plan

# Execute
/implement-plan

Quick Command Categories

Get Started (First Day)

/onboard                        # Guided onboarding
/hello # Test installation
/verify-installation # Verify setup
/orient # Session orientation

Find Things

/what                           # Component inventory
/which <task> # Agent recommendation
/how <task> # How-to guidance
/where <component> # File location

Build Things

/project-new <name>             # New project
/implement <feature> # Implement feature
/prototype <idea> # Quick prototype
/oneshot <feature> # Single-shot implementation

Review Things

/analyze                        # Code analysis
/moe-judges <target> # Multi-expert evaluation
/full-review # Comprehensive review
/local-review # Quick local review

Git Operations

/commit                         # Create commits
/git-sync # Sync branches
/describe-pr # PR description
/smart-merge # Resolve conflicts

Documentation

/document                       # Generate docs
/readme-gen # Generate README
/explain <code> # Explain code
/doc-generate # Full documentation

Quality & Security

/security-scan                  # Security scanning
/test-generate # Generate tests
/test-coverage # Coverage analysis
/production-audit # Production readiness

Command Execution Rules

All CODITECT commands follow these rules:

  1. IMMEDIATE EXECUTION - Commands execute when invoked, no questions first
  2. FULL OUTPUT - Always shows complete tool output
  3. SUMMARY PROVIDED - Always provides summary after completion
  4. USER INVOCATION IS CONFIRMATION - No additional confirmation needed (except deletions, force-push, production deploys)

This means:

/commit          # Executes immediately, doesn't ask "are you sure?"
/moe-judges X # Assembles panel and evaluates immediately
/git-sync # Syncs immediately

Finding More Commands

List All Commands

/what commands                  # All 192 commands
/what commands --search "git" # Search commands
ls ~/.coditect/commands/ # File listing

Get Command Help

/what is /commit                # Full command details
/how use /moe-judges # Usage guidance

Browse by Category

# In ~/.coditect/commands/
ls ~/.coditect/commands/ | grep -i git # Git commands
ls ~/.coditect/commands/ | grep -i moe # MoE commands
ls ~/.coditect/commands/ | grep -i test # Testing commands

Summary: 10 Commands to Master

#CommandMaster This For
1/orientSession starts
2/whichAgent discovery
3/agentTask execution
4/whatComponent discovery
5/moe-judgesQuality evaluation
6/commitGit commits
7/cxqMemory search
8/cxMemory save
9/project-newNew projects
10/researchVerify assumptions

Next Steps


Commands: 192 | Agents: 157 | Total Components: 2,172 Last Updated: January 12, 2026 Framework: CODITECT v2.0.0