lev-cdo-router

name	lev-cdo-router
description	[WHAT] Classification engine and complexity-based routing for CDO workflows [HOW] Analyzes input intent, confidence, complexity → routes to Quick/Base/Deep/Epic [WHEN] First step of every CDO invocation - determines execution strategy [WHY] Right-sized workflows prevent over/under-engineering Triggers: Called automatically by lev-cdo main skill
skill_type	playbook
category	process-thinking
lifecycle_integration	{"stage":"crystallizing","input_artifact":"raw query or problem statement","output_artifact":"classification metadata + routing decision"}
related_skills	["skill://lev-cdo/workflows","skill://lev-find","skill://work"]
plankton	false

CDO Router - Classification & Routing

Lev Concept

What is this? The CDO router is the entry point for all graph-based thinking workflows. It analyzes input to determine intent, confidence level, and complexity, then routes to the appropriate execution pattern.

Why does it exist? Without classification, every problem gets the same heavyweight treatment. The router ensures quick questions get quick answers, while complex problems get multi-agent graph exploration.

When to use it:

• Automatically called by lev cdo command
• Can be invoked standalone for classification-only: lev cdo classify "{input}"

---

CLI Commands

Primary Command

# Auto-classify and route (called by parent skill)
lev cdo classify "{input}"

Output Format

classify:
  intent: "{question|command|idea|status|research}"
  confidence: 0.0-1.0
  complexity: "{quick|base|deep|epic}"
  tags: [auto-detected, keywords]
  capture: "{bd|idea|scratchpad|none}"
  route_to: "{workflows/quick|workflows/base|workflows/deep|workflows/epic}"

---

Workflows

Workflow 1: Classification

Use case: Every CDO invocation starts here

Steps:

1. Parse input for intent signals
2. Assess confidence (how well-understood is the problem?)
3. Determine complexity (scope, interdependencies, unknowns)
4. Tag with keywords for context
5. Decide capture destination (BD epic, idea file, scratch, none)
6. Route to appropriate workflow sub-skill

Input: Raw user query or problem statement Output: Classification metadata + routing decision

Example:

# Input: "Should we use GraphQL or REST?"
# Output:
classify:
  intent: question
  confidence: 0.85
  complexity: base
  tags: [api, architecture, design-decision]
  capture: idea
  route_to: workflows/base

---

Workflow 2: DoR Validation

Use case: Before executing Deep or Epic workflows

Steps:

1. Sanity Check - Verify problem/solution alignment
2. Prior Art Check - Search for existing implementations
3. Architecture Compliance - Validate YAML-first, LLM-first principles
4. Scope Validation - User approval for high-effort work

Input: Classification metadata + user query Output: PASS/FAIL + validation notes

Example:

# Complexity: deep
# Triggers: Sanity check + prior art + architecture compliance

DoR Validation:
  sanity_check: PASS
  prior_art: FOUND (3 existing patterns in ~/lev/ideas/)
  architecture_compliance: PASS
  scope_validation: PENDING (requires user approval for 500+ LOC)
  verdict: BLOCKED until scope approved

---

Classification Logic

Intent Detection

Pattern	Intent
"how", "what", "why", "when"	question
"do X", "create Y", "implement Z"	command
"we should", "what if", "consider"	idea
"show status", "what's blocked"	status
"find out", "investigate", "explore"	research

Confidence Scoring

confidence_rules:
  - known_solution_exists: +0.30
  - clear_requirements: +0.20
  - prior_art_found: +0.15
  - single_domain: +0.10
  - vague_requirements: -0.20
  - unknown_domain: -0.30
  - conflicting_constraints: -0.25

Complexity Scoring

complexity_factors:
  quick:  # 0-2 factors
    - Single question
    - Known answer pattern
    - No dependencies

  base:   # 3-5 factors
    - Multiple perspectives needed
    - 2-3 related domains
    - Some prior art exists

  deep:   # 6-9 factors
    - Root cause unknown
    - Multiple hypotheses
    - Cross-domain analysis
    - 5+ agents needed

  epic:   # 10+ factors
    - Strategic decision
    - Multi-session work
    - Unknown unknowns
    - BD epic required

---

DoR Validation Gates

Gate 1: Sanity Check (Deep/Epic only)

Questions:

• What's the ACTUAL problem being solved?
• What's the simplest solution?
• Is this a 10-line fix or 1000-line system?
• Does the request match the solution complexity?

Validation:

checks:
  - estimated_loc > 500 AND request_length < 100: SCOPE_MISMATCH
  - user_says "simple|just|quick": EXPECTATION_MISMATCH
  - solution_simpler_than_problem: OVER_ENGINEERING

Action on FAIL: Use AskUserQuestion to clarify scope

---

Gate 2: Prior Art Check (ALWAYS)

Search locations:

# Codebase search
lev get "{keywords}" --indexes codebase

# Documentation search
lev get "{keywords}" --indexes documentation

# BD beads search
bd list --title-contains "{keywords}"
bd list --desc-contains "{keywords}"

# PM artifacts search
find ~/.lev/pm/handoffs -name "*{keywords}*.md"
find ~/.lev/pm/designs -name "*{keywords}*.md"

Required output:

• Existing implementations found: [yes/no]
• Can we reuse/compose: [yes/no]
• Why we're NOT reusing: [reason if applicable]

Action on FOUND: Propose reuse/composition before new implementation

---

Gate 3: Architecture Compliance (Deep/Epic design)

YAML-first, LLM-first principles:

Questions:

• Is behavior in YAML or hardcoded?
• Can users customize without code changes?
• Where's the LLM in this design?
• If using FlowMind: YAML compiler or runtime consumer?

Anti-patterns to catch:

violations:
  - hardcoded_routing: "if factorCount <= 2 then quick else deep"
  - pattern_matching: "regex instead of LLM reasoning"
  - locked_behavior: "users can only tweak threshold numbers"
  - flowmind_misuse: "LLM analyzes → hardcoded if/else executes"

Required:

• Behavior defined in YAML (customizable)
• LLM reasoning drives decisions
• FlowMind compiles YAML → behavior (not just analysis)

Action on FAIL: Reject design, propose YAML-first refactor

---

Gate 4: Scope Validation (Deep/Epic only)

Triggers:

• Estimated effort > 4 hours
• LOC > 500
• Multi-session work required
• BD epic recommended

Action:

1. Present effort estimate to user
2. Confirm BD epic creation
3. Require justification if NOT reusing existing solutions
4. Get explicit approval before proceeding

---

Message Footer Protocol

Every CDO response ends with structured footer:

---
🆔 **Session:** {session_id} | 📋 {status} | 🏷️ [{tags}]

**Synths Active:** [{running_ephemeral_workflows}]

**Queued:**
- [ ] {pending_bd_update}
- [ ] {pending_idea_capture}

**Confirmed:** [x] {just_committed}

🪄 **Next:**
1. {action_verb} {target} [skill://...]
2. {action_verb} {target} [skill://...]
3. All | 4. ⬅️ Back
---

Variables:

Variable	Source	Example
session_id	Hook/UUID	69C2-1408
status	Classification	crystallizing
tags	Auto-detect	[cdo, planning]
synths	Active loops	[research-spike, design-poc]
queued	Pending commits	BD/idea updates
next_steps	Context-generated	Skill invocations

---

Routing Decision Table

Confidence	Complexity	Route To	Notes
≥0.90	quick	workflows/quick	Direct execution, 1-2 agents
≥0.80	base	workflows/base	Fan-out/merge, 2-3 agents
≥0.60	deep	workflows/deep	Multi-turn chains, 3-5 agents
<0.60	epic	workflows/epic	BD-tracked, 5+ agents, multi-session

Override: User can force complexity level with --complexity={level} flag

---

Relates

Depends On

• skill://lev-find - Prior art search (legacy skill name for lev get backend)
• skill://bd - Epic tracking for complex work

Works With

• skill://lev-cdo/workflows - Receives routing decisions
• skill://work - Architecture/lifecycle compliance validation (ALIGN gate)

Blocks/Enables

• Blocks: Execution until DoR passes
• Enables: Right-sized workflow execution

Lifecycle Position

• Before: Raw user input
• After: Workflow execution (quick/base/deep/epic)
• Parallel: None (sequential gate)

Technique Map

• Role definition - Clarifies operating scope and prevents ambiguous execution.
• Context enrichment - Captures required inputs before actions.
• Output structuring - Standardizes deliverables for consistent reuse.
• Step-by-step workflow - Reduces errors by making execution order explicit.
• Edge-case handling - Documents safe fallbacks when assumptions fail.

Technique Notes

These techniques improve reliability by making intent, inputs, outputs, and fallback paths explicit. Keep this section concise and additive so existing domain guidance remains primary.

Prompt Architect Overlay

Role Definition

You are the prompt-architect-enhanced specialist for router, responsible for deterministic execution of this skill's guidance while preserving existing workflow and constraints.

Input Contract

• Required: clear user intent and relevant context for this skill.
• Preferred: repository/project constraints, existing artifacts, and success criteria.
• If context is missing, ask focused questions before proceeding.

Output Contract

• Provide structured, actionable outputs aligned to this skill's existing format.
• Include assumptions and next steps when appropriate.
• Preserve compatibility with existing sections and related skills.

Edge Cases & Fallbacks

• If prerequisites are missing, provide a minimal safe path and request missing inputs.
• If scope is ambiguous, narrow to the highest-confidence sub-task.
• If a requested action conflicts with existing constraints, explain and offer compliant alternatives.