name	router
description	Intelligent workflow selection that analyzes user requests and recommends the optimal ModelChorus workflow with parameters

ROUTER

Overview

The ROUTER skill helps AI agents make intelligent decisions about which ModelChorus workflow to invoke for a given user request. Instead of guessing or defaulting to CHAT for everything, this skill provides a systematic analysis that considers the user's intent, problem complexity, constraints, and desired outcomes.

Available Workflows:

CHAT - Single-model conversation with threading
CONSENSUS - Multi-model parallel consultation with synthesis
THINKDEEP - Extended reasoning with hypothesis tracking
ARGUMENT - Three-role dialectical reasoning (Creator/Skeptic/Moderator)
IDEATE - Creative brainstorming with quantity/creativity control

When to Use This Skill:

The main AI agent should invoke this skill when:

User makes a request that could map to multiple workflows
Uncertain which workflow would best serve the user's needs
Need to extract appropriate parameters for the selected workflow
Want to suggest workflow sequences for complex multi-step tasks

This skill outputs a structured decision including:

Primary workflow recommendation
Confidence level (high/medium/low)
Workflow-specific parameters
Alternative workflows (when applicable)
Suggested sequences for multi-step tasks
Clarifying questions (when confidence is low)

Signal Extraction Guide

Before selecting a workflow, extract these signals from the user's request:

1. Primary Goal (The Core Intent)

Goal Category	Keywords/Patterns	Maps To
Answer/Opinion	"what is", "explain", "how does", "why", "quick question"	CHAT
Evaluation	"evaluate", "critique", "review", "pros and cons", "pre-mortem", "assess"	ARGUMENT, CONSENSUS
Generation	"brainstorm", "ideas for", "suggest ways", "creative solutions", "come up with"	IDEATE
Investigation	"debug", "investigate", "root cause", "why is X happening", "analyze problem"	THINKDEEP, CHAT
Decision	"choose between", "compare A vs B", "decide on", "select", "which should"	CONSENSUS, ARGUMENT

2. Complexity & Scope Indicators

Complexity Level	Signals	Workflow Implications
Simple	"quick", "simple", "straightforward", factual questions	Favors CHAT
Moderate	Standard feature requests, comparisons, evaluations	ARGUMENT, IDEATE, RESEARCH
Complex	"intermittent", "ambiguous", "multi-step", "complex system", "race condition", "unclear cause"	Strongly favors THINKDEEP

3. Perspective Requirements

Requirement	Signals	Workflow
Single perspective sufficient	No explicit need for validation	CHAT, THINKDEEP, ARGUMENT, IDEATE
Multiple perspectives needed	"validate", "cross-check", "multiple viewpoints", "blind spots", "consensus", "compare approaches"	CONSENSUS

4. Decision Impact & Risk

Impact Level	Signals	Workflow
High stakes	"architecture", "production", "security review", "major decision", "high-impact", "critical"	CONSENSUS (for validation)
Moderate stakes	Standard feature decisions, design choices	ARGUMENT, CHAT
Low stakes	Exploratory questions, learning	CHAT, IDEATE

5. Output Requirements

Requirement	Signals	Workflow
Multiple options	"different ways", "explore possibilities", "alternatives", "various approaches"	IDEATE, CONSENSUS
Structured debate	"pro and con", "strengths and weaknesses", "balanced view", "critique"	ARGUMENT
Systematic analysis	"step by step", "methodical", "build evidence", "track progress"	THINKDEEP

6. Constraints

Constraint Type	Signals	Impact
Time limited	"quick", "fast", "urgent", "in a hurry", "ASAP"	Favor CHAT, IDEATE, ARGUMENT; avoid THINKDEEP, CONSENSUS
Budget limited	"cheap", "low-cost", "economical", "minimal expense"	Favor CHAT, IDEATE; avoid CONSENSUS
Quality priority	"thorough", "comprehensive", "validated", "high quality", "certain"	Favor CONSENSUS, THINKDEEP

7. Interaction Style

Style	Signals	Workflow
Conversational/iterative	"let's explore", "I'll have follow-ups", "iterative", ongoing discussion	CHAT (supports threading)
One-shot synthesis	"give me a complete analysis", "comprehensive answer"	CONSENSUS, ARGUMENT
Multi-step investigation	"investigate step by step", "build confidence", "test hypotheses"	THINKDEEP (supports state persistence)

8. Artifact to Analyze

Presence	Signals	Workflow
Specific proposal/design provided	User includes code, design doc, architecture diagram, or explicit proposal to evaluate	Strong signal for ARGUMENT
No artifact	General question or request	Other workflows

Routing Procedure

Follow this 6-step procedure to select the optimal workflow:

Step 0: Parse Constraints and Context

Extract immediately:

Budget constraints (limited/flexible)
Time constraints (urgent/flexible)
Quality requirements (standard/high)
Conversation continuation need (threading expected?)

References:

WORKFLOW_SELECTION_GUIDE.md lines 445-468 ("By Constraints")
CHAT supports threading (line 65)
CONSENSUS does not support threading (line 110)
THINKDEEP supports state persistence (lines 143-145)

Step 1: Identify Primary Goal and Complexity

Determine:

What is the user's core intent? (Answer/Ideas/Evaluation/Investigation/Decision/Information)
What is the problem complexity? (Simple/Moderate/Complex)

References:

WORKFLOW_SELECTION_GUIDE.md line 565 ("What's my goal?")
Line 436-442 ("By Problem Type" table)

Step 2: Apply Prioritized Rules (Highest to Lowest Precedence)

Apply these rules in order. First match wins unless overridden by constraints in Step 3:

Specific proposal/decision to evaluate (artifact provided) → ARGUMENT
- Reference: Lines 193-199 ("Evaluating a specific proposal")
Creative options/brainstorming needed → IDEATE
- Reference: Lines 232-238 ("Need creative options")
High-impact decision needing multiple perspectives → CONSENSUS
- Reference: Lines 113-119 ("Decision has significant impact... Multiple perspectives")
Complex/ambiguous investigation (multi-step reasoning) → THINKDEEP
- Reference: Lines 153-159 ("Problem is complex... Multi-step analysis beneficial")
Otherwise (default) → CHAT
- Reference: Lines 74-80 ("Quick answer... One model's perspective")

###Step 3: Apply Constraint-Based Overrides

If time/budget limited:

Demote CONSENSUS and THINKDEEP
Promote CHAT, IDEATE, ARGUMENT
Reference: Lines 445-459 ("When budget/time is limited")

If quality is priority:

Promote CONSENSUS, THINKDEEP
Reference: Lines 461-468 ("When quality is priority")

Step 4: Conversation-Mode Routing

If user expects multi-turn conversation:

Prefer CHAT or THINKDEEP (both support threading/persistence)
Avoid CONSENSUS (no conversation continuation)
Reference: Line 65 (CHAT threading), Line 110 (CONSENSUS no threading), Lines 143-145 (THINKDEEP persistence)

If user asks to "validate" or "cross-check" after investigation:

Suggest CONSENSUS as next step
Reference: Lines 491-497 ("Investigate → Validate Pattern")

Step 5: Sequence Suggestion

If multiple intents detected or staged work identified:

Use recommended patterns from WORKFLOW_SELECTION_GUIDE.md lines 475-505:

Generate → Evaluate Pattern: IDEATE → ARGUMENT → CONSENSUS
- Use for: Feature planning, solution design
Investigate → Validate Pattern: THINKDEEP → CONSENSUS → ARGUMENT
- Use for: Complex debugging, root cause analysis
Quick → Deep Pattern: CHAT → THINKDEEP → CONSENSUS
- Use for: Exploratory analysis that becomes complex

Step 6: Low-Confidence Routing

If confidence is low or request is ambiguous:

Formulate ONE targeted clarifying question
Provide a provisional default (usually CHAT for safety)
Include rationale for why clarification is needed

Examples of ambiguous requests:

"Help with my project" → Ask: "What kind of help? Generate ideas, debug an issue, or evaluate a plan?"
"Look into this API" → Ask: "Should I investigate a specific problem, research best practices, or review the design?"

Anti-Pattern Prevention

DO NOT route to these workflows in these situations:

❌ CONSENSUS for Simple Questions

Bad: "What's the capital of France?" → CONSENSUS
Why wrong: Overkill, wastes resources
Use instead: CHAT
Reference: Lines 511-515

❌ CHAT for Complex Investigations

Bad: "Debug this intermittent race condition" → CHAT
Why wrong: No systematic investigation, no hypothesis tracking
Use instead: THINKDEEP
Reference: Lines 516-520

❌ IDEATE for Evaluation

Bad: "Is this database schema design good?" → IDEATE
Why wrong: IDEATE generates, doesn't evaluate
Use instead: ARGUMENT or CONSENSUS
Reference: Lines 521-525

❌ THINKDEEP for Quick Facts

Bad: "What HTTP status code means unauthorized?" → THINKDEEP
Why wrong: Overly complex for simple lookup
Use instead: CHAT
Reference: Lines 531-535

Workflow-Specific Parameter Suggestions

When you select a workflow, suggest appropriate parameters based on the user request:

CHAT Parameters

continue: thread ID if user is in ongoing conversation
provider: claude (default), gemini, codex, cursor-agent

CONSENSUS Parameters

strategy: "synthesize" (default for structured view), "all_responses" (to see each model separately), "majority" (factual queries), "weighted" (most detailed), "first_valid" (speed priority)
provider: List of 2-4 providers (default: ["claude", "gemini"])
Reference: Lines 308, 420

THINKDEEP Parameters

step: Brief description of what to investigate in this step
step_number: 1 (for new investigations)
total_steps: Estimate based on complexity (1-3 for simple, 3-5 for moderate, 6-10 for complex)
next_step_required: true (for multi-step), false (single step)
findings: Initial observations or "Starting investigation"
confidence: "exploring" (default for new), or carry forward from previous
Reference: Lines 340-341

ARGUMENT Parameters

No specific parameters beyond the prompt

IDEATE Parameters

num_ideas: 3-4 (quick), 5-7 (standard), 8-10 (comprehensive), 10+ (exhaustive)
Reference: Line 412

Output Format Specification

Return a JSON-structured decision object:

{
  "workflow": "CONSENSUS",
  "confidence": "high",
  "rationale": "User requests comparative evaluation for high-impact architecture decision. Multiple perspectives reduce blind spots and validate choice.",
  "parameters": {
    "strategy": "synthesize",
    "provider": ["claude", "gemini", "codex"]
  },
  "alternatives": [
    {
      "workflow": "ARGUMENT",
      "when_to_use": "If budget/time constraints tighten",
      "reason": "Faster structured critique of top option, single model"
    }
  ],
  "sequence": [
    {
      "step": 1,
      "workflow": "CONSENSUS",
      "trigger": "Initial evaluation"
    },
    {
      "step": 2,
      "workflow": "ARGUMENT",
      "trigger": "After short-listing top choice, critique it"
    }
  ],
  "clarifying_question": null,
  "used_signals": [
    "compare X vs Y",
    "high-impact decision",
    "architecture choice",
    "need validation"
  ]
}

Field Descriptions

Field	Type	Required	Description
`workflow`	string	Yes	One of: CHAT, CONSENSUS, THINKDEEP, ARGUMENT, IDEATE
`confidence`	string	Yes	One of: "high", "medium", "low"
`rationale`	string	Yes	1-2 sentences explaining why this workflow was selected
`parameters`	object	No	Workflow-specific parameters extracted from request
`alternatives`	array	No	Other viable workflows with context for when to use them
`sequence`	array	No	Multi-step workflow plan for complex tasks
`clarifying_question`	string	No	Single targeted question when confidence is low
`used_signals`	array	Yes	List of extracted signals for transparency/debugging

Example Routing Decisions

Example 1: Simple Question → CHAT

User Request: "What is JWT authentication?"

Decision:

{
  "workflow": "CHAT",
  "confidence": "high",
  "rationale": "Simple factual question requiring quick explanation. Single perspective sufficient.",
  "parameters": {},
  "used_signals": ["what is", "simple question", "factual"]
}

Example 2: High-Impact Decision → CONSENSUS

User Request: "Choose between PostgreSQL, MongoDB, and DynamoDB for our e-commerce platform (10k users, ACID required, team has SQL background)"

Decision:

{
  "workflow": "CONSENSUS",
  "confidence": "high",
  "rationale": "High-impact technology selection with significant implications. Multiple model perspectives provide validation and reduce blind spots.",
  "parameters": {
    "strategy": "synthesize",
    "provider": ["claude", "gemini", "codex"]
  },
  "sequence": [
    {
      "step": 1,
      "workflow": "CONSENSUS",
      "trigger": "Make final decision with synthesis"
    },
    {
      "step": 2,
      "workflow": "ARGUMENT",
      "trigger": "Critique the selected choice"
    }
  ],
  "used_signals": [
    "choose between",
    "high-impact",
    "e-commerce platform",
    "need validation",
    "comparison"
  ]
}

Example 3: Complex Debugging → THINKDEEP

User Request: "API latency increased from 100ms to 2s after deployment. Intermittent, affects all endpoints."

Decision:

{
  "workflow": "THINKDEEP",
  "confidence": "high",
  "rationale": "Complex investigation requiring systematic hypothesis testing. Multi-step analysis needed to build confidence in root cause.",
  "parameters": {
    "step": "Investigate 20x latency increase after deployment",
    "step_number": 1,
    "total_steps": 4,
    "next_step_required": true,
    "findings": "Latency affects all endpoints equally, started immediately after 3pm deployment",
    "confidence": "exploring"
  },
  "sequence": [
    {
      "step": 1,
      "workflow": "THINKDEEP",
      "trigger": "Systematic investigation"
    },
    {
      "step": 2,
      "workflow": "CONSENSUS",
      "trigger": "After identifying likely cause, validate findings"
    }
  ],
  "used_signals": [
    "investigate",
    "complex problem",
    "intermittent",
    "root cause needed",
    "multi-step analysis"
  ]
}

Example 4: Proposal Evaluation → ARGUMENT

User Request: "Evaluate this API design: [paste design]. Focus on scalability, security, and developer experience."

Decision:

{
  "workflow": "ARGUMENT",
  "confidence": "high",
  "rationale": "Specific proposal provided for evaluation. Structured dialectical critique (Creator/Skeptic/Moderator) provides balanced analysis.",
  "parameters": {},
  "alternatives": [
    {
      "workflow": "CONSENSUS",
      "when_to_use": "If you need multiple real model perspectives",
      "reason": "More thorough but higher cost"
    }
  ],
  "used_signals": [
    "evaluate",
    "specific proposal provided",
    "critique needed",
    "focus on specific criteria"
  ]
}

Example 5: Creative Brainstorming → IDEATE

User Request: "Brainstorm creative ways to improve user onboarding. Current drop-off is 40%."

Decision:

{
  "workflow": "IDEATE",
  "confidence": "high",
  "rationale": "Creative idea generation needed. IDEATE provides structured brainstorming with configurable quantity.",
  "parameters": {
    "num_ideas": 6
  },
  "sequence": [
    {
      "step": 1,
      "workflow": "IDEATE",
      "trigger": "Generate creative ideas"
    },
    {
      "step": 2,
      "workflow": "ARGUMENT",
      "trigger": "After selecting top 2-3 ideas, evaluate each"
    }
  ],
  "used_signals": [
    "brainstorm",
    "creative ways",
    "generate ideas",
    "improve onboarding"
  ]
}

Example 6: Ambiguous Request (Low Confidence) → Clarifying Question

User Request: "Help with my project"

Decision:

{
  "workflow": "CHAT",
  "confidence": "low",
  "rationale": "Request is too vague to determine optimal workflow. Defaulting to CHAT as safe fallback but requesting clarification.",
  "clarifying_question": "What kind of help do you need? Are you trying to:\n- Generate creative ideas (brainstorming)\n- Debug or investigate a problem\n- Evaluate or critique a proposal\n- Get a quick answer or opinion",
  "used_signals": [
    "ambiguous intent",
    "no clear goal",
    "generic request"
  ]
}

Example 7: Multiple Valid Workflows → Alternatives Provided

User Request: "Review this database schema design"

Decision:

{
  "workflow": "ARGUMENT",
  "confidence": "medium",
  "rationale": "Evaluation of specific design. ARGUMENT provides structured critique (Creator/Skeptic/Moderator) at moderate cost.",
  "parameters": {},
  "alternatives": [
    {
      "workflow": "CONSENSUS",
      "when_to_use": "If you want multiple real model perspectives for higher confidence",
      "reason": "More thorough validation but higher cost"
    },
    {
      "workflow": "CHAT",
      "when_to_use": "If budget/time is very limited",
      "reason": "Quick single-perspective review"
    }
  ],
  "used_signals": [
    "review",
    "specific artifact (schema)",
    "evaluation needed",
    "multiple valid approaches"
  ]
}

Example 8: Time Constraint Override → CHAT

User Request: "Quick: Should we use microservices or monolith for our 5-person team?"

Decision:

{
  "workflow": "CHAT",
  "confidence": "medium",
  "rationale": "Time constraint ('quick') overrides typical CONSENSUS recommendation for architecture decisions. CHAT provides fast guidance with note about more thorough options.",
  "parameters": {},
  "alternatives": [
    {
      "workflow": "CONSENSUS",
      "when_to_use": "If time permits and decision impact justifies it",
      "reason": "Multi-model validation for high-impact architecture choice"
    }
  ],
  "used_signals": [
    "quick (time constraint)",
    "architecture decision",
    "constraint override applied"
  ]
}

Example 9: Sequence Suggestion → Multi-Step Plan

User Request: "We need to add a new payment provider. Help us choose and implement."

Decision:

{
  "workflow": "CONSENSUS",
  "confidence": "high",
  "rationale": "Multi-step task detected: evaluate options → critique → implement. Starting with CONSENSUS to gather multiple perspectives on payment providers.",
  "parameters": {
    "strategy": "synthesize",
    "provider": ["claude", "gemini", "codex"]
  },
  "sequence": [
    {
      "step": 1,
      "workflow": "CONSENSUS",
      "trigger": "Evaluate top 2-3 payment provider options with multiple perspectives"
    },
    {
      "step": 2,
      "workflow": "ARGUMENT",
      "trigger": "Critique selected provider before implementation"
    },
    {
      "step": 3,
      "workflow": "CHAT",
      "trigger": "Implementation guidance and iterative support"
    }
  ],
  "used_signals": [
    "multi-step task",
    "decision + implementation",
    "payment provider selection"
  ]
}

Technical Contract

Input

Required:

user_request (string): The user's full request/question

Optional:

conversation_context (object): Information about ongoing conversation
- has_thread: boolean indicating if user is in threaded conversation
- previous_workflow: string indicating last workflow used
explicit_constraints (object): User-specified constraints
- time_limited: boolean
- budget_limited: boolean
- quality_priority: boolean

Output

Returns a JSON object with the structure defined in "Output Format Specification" section.

Required fields:

workflow: string
confidence: string (high/medium/low)
rationale: string
used_signals: array of strings

Optional fields:

parameters: object (workflow-specific)
alternatives: array of objects
sequence: array of objects
clarifying_question: string

Clarifying Question Templates

Use these templates when confidence is low:

Goal Ambiguity

Template: "What's your primary goal? Are you trying to:\n- [Option A related to request]\n- [Option B related to request]\n- [Option C related to request]"

Example: "What's your primary goal? Are you trying to:\n- Generate creative ideas (brainstorming)\n- Evaluate an existing proposal\n- Get a quick answer or opinion"

Complexity Unclear

Template: "How deep should I go? Do you need:\n- Quick guidance (fast, single perspective)\n- Thorough analysis (systematic investigation)\n- Validated decision (multiple perspectives)"

Multiple Options Unclear

Template: "Should I:\n- Explore multiple options/ideas\n- Evaluate a specific proposal\n- Give you one recommendation"

Best Practices for Using This Skill

Always extract signals first - Don't jump to conclusions; systematically analyze the request
Consider constraints early - Time/budget/quality requirements can override default selections
Provide alternatives for medium confidence - Help users understand trade-offs
Suggest sequences for complex tasks - Guide users through multi-step workflows
Default to CHAT when uncertain - It's the safest, most flexible option
Be explicit in rationale - Reference specific signals and guide sections
Pre-populate parameters - Save users time by suggesting sensible defaults
Ask ONE clarifying question - Don't overwhelm with multiple questions; focus on the decision boundary

This skill is the intelligent decision layer for ModelChorus. Use it to ensure users get routed to the optimal workflow for their specific needs.

Install Skill

SKILL.md

ROUTER

Overview

Signal Extraction Guide

1. Primary Goal (The Core Intent)

2. Complexity & Scope Indicators

3. Perspective Requirements

4. Decision Impact & Risk

5. Output Requirements

6. Constraints

7. Interaction Style

8. Artifact to Analyze

Routing Procedure

Step 0: Parse Constraints and Context

Step 1: Identify Primary Goal and Complexity

Step 2: Apply Prioritized Rules (Highest to Lowest Precedence)

Step 4: Conversation-Mode Routing

Step 5: Sequence Suggestion

Step 6: Low-Confidence Routing

Anti-Pattern Prevention

❌ CONSENSUS for Simple Questions

❌ CHAT for Complex Investigations

❌ IDEATE for Evaluation

❌ THINKDEEP for Quick Facts

Workflow-Specific Parameter Suggestions

CHAT Parameters

CONSENSUS Parameters

THINKDEEP Parameters

ARGUMENT Parameters

IDEATE Parameters

Output Format Specification

Field Descriptions

Example Routing Decisions

Example 1: Simple Question → CHAT

Example 2: High-Impact Decision → CONSENSUS

Example 3: Complex Debugging → THINKDEEP

Example 4: Proposal Evaluation → ARGUMENT

Example 5: Creative Brainstorming → IDEATE

Example 6: Ambiguous Request (Low Confidence) → Clarifying Question

Example 7: Multiple Valid Workflows → Alternatives Provided

Example 8: Time Constraint Override → CHAT

Example 9: Sequence Suggestion → Multi-Step Plan

Technical Contract

Input

Output

Clarifying Question Templates

Goal Ambiguity

Complexity Unclear

Multiple Options Unclear

Best Practices for Using This Skill