lev-cdo - Graph-Based Agentic Thinking

Lev Concept

What is this? CDO (Complexity-Driven Orchestration) is a graph-based execution system where agents read/write to disk in structured workflows. Problems are classified by complexity, then routed to appropriate patterns (Quick/Base/Deep/Epic).

Why does it exist? Traditional linear thinking fails for interconnected problems. CDO prevents groupthink by ensuring agents never see each other's work during execution - only the final synthesis step reads all artifacts together.

When to use it:

• Design work - Architecture decisions, system design
• Complex debugging - Root cause analysis with multiple hypotheses
• Strategic analysis - Multi-perspective evaluation of decisions
• Assumption exploration - Socratic questioning of beliefs
• Research synthesis - Integrating findings from multiple sources

When NOT to use it:

• Simple questions with known answers → Use Quick workflow or just answer
• Single perspective sufficient → Direct execution
• No complexity/interdependencies → Standard task execution

---

CLI Commands

Primary Commands

# Auto-classify and route to appropriate workflow
lev cdo "{query or problem statement}"

# Force specific complexity level
lev cdo --complexity=quick "{query}"
lev cdo --complexity=base "{query}"
lev cdo --complexity=deep "{query}"
lev cdo --complexity=epic --bd-epic=clawd-xxx "{query}"

# Classification only (no execution)
lev cdo classify "{input}"

# Debug mode (RCA workflow)
lev cdo debug "{error description}"

Examples

# Example 1: Auto-classify (likely: base)
lev cdo "Should we use GraphQL or REST for the API?"

# Example 2: Force deep complexity
lev cdo --complexity=deep "Explore assumptions behind microservices"

# Example 3: Debug workflow
lev cdo debug "Gateway crashes on Telegram messages"

# Example 4: Epic with BD tracking
lev cdo --complexity=epic --bd-epic=clawd-042 "Evaluate voice assistant architecture"

---

Workflows

Overview: Four Complexity Levels

Level	Agents	Turns	Pattern	Use When
Quick	1-2	1	Sequential	Simple question, single perspective
Base	2-3	2	Fan-out/merge	Need 2+ perspectives, then synthesis
Deep	3-5	3-5	Multi-turn chains	Root cause analysis, assumption drilling
Epic	5+	5-10	BD-tracked phases	Strategic analysis, multi-session work

Routing logic:

Confidence ≥0.90 → Quick (direct execution)
Confidence ≥0.80 → Base (fan-out perspectives → synthesis)
Confidence ≥0.60 → Deep (multi-turn chains with convergence)
Confidence <0.60 → Epic (BD-tracked, human-in-loop)

---

Workflow 1: Classification & Routing

Use case: First step of every CDO invocation

Handled by: skill://lev-cdo/router

Process:

1. Parse input for intent (question, command, idea, research)
2. Assess confidence (0.0-1.0)
3. Determine complexity (quick, base, deep, epic)
4. Validate DoR (Definition of Ready)
5. Route to appropriate workflow

DoR Gates:

• Sanity check (problem/solution alignment)
• Prior art check (search for existing implementations)
• Architecture compliance (YAML-first, LLM-first principles)
• Scope validation (user approval for high-effort work)

See: lev-cdo/router/SKILL.md for full details

---

Workflow 2: Quick (1-2 agents, 1 turn)

Example: "Summarize this document"

Execution:

Agent: Read 00-input.md → Write FINAL-summary.md

Convergence: Single output, no iteration

---

Workflow 3: Base (2-3 agents, 2 turns)

Example: "Analyze pros and cons of X"

Pattern: Fan-out (parallel perspectives) → Merge (synthesis)

Execution:

Turn 1 (parallel):
  Agent A: Argue FOR → 01a-pros.md
  Agent B: Argue AGAINST → 01b-cons.md

Turn 2:
  Agent C: Synthesize → FINAL-analysis.md

Convergence: All perspectives collected + synthesis complete

---

Workflow 4: Deep (3-5 agents, 3-5 turns)

Example: "Explore foundational assumptions behind belief X"

Pattern: Multi-turn chains (socratic drilling, debate, synthesis)

Execution:

Turn 1: Paraphrase belief clearly
Turn 2: Steelman (strengthen before critique)
Turn 3: Dig axioms (3-level socratic drilling)
Turn 4: Map elements (extract components)
Turn 5: Multi-devils debate (FOR/AGAINST/INTEGRATE)
Turn 6: Synthesize findings
Turn 7: Reflection loop (meta-analysis)

Convergence: All turns complete + confidence ≥0.80

See: lev-cdo/workflows/SKILL.md for full examples

---

Workflow 5: Epic (5+ agents, 5-10 turns, BD-tracked)

Example: "Should we bet on technology X? Full strategic analysis."

Pattern: BD-tracked phases with dependencies

Execution:

Phase 1: Research (ergodicity, lindy-effect, cynefin)
  → BD: clawd-xxx.1

Phase 2: Analysis (dig-axioms, morphological, TRIZ-40)
  → BD: clawd-xxx.2 (depends on Phase 1)

Phase 3: Synthesis (resonance loop until convergence)
  → BD: clawd-xxx.3 (depends on Phase 2)

Phase 4: Output (synthesize-apply, reflection-loop)
  → BD: clawd-xxx.4 (depends on Phase 3)

Convergence: All BD milestones closed + final report approved

See: lev-cdo/workflows/SKILL.md for Epic patterns

---

Workflow 6: Debug RCA (7-step systematic debugging)

Triggered by: "uld", "ultradebug" keywords

Handled by: skill://lev-cdo/debug

Steps:

1. REPRODUCE - Define exact failure
2. ISOLATE - Narrow to minimal case
3. TRACE - Find call path + working code
4. HYPOTHESIZE - Form 2-3 theories
5. VERIFY - Test hypotheses with evidence
6. FIX - Apply minimal fix (NO scope creep)
7. VALIDATE - Adversarial validation (ralph)

Convergence: All 7 steps complete + PASS verdict

See: lev-cdo/debug/SKILL.md for full RCA workflow

---

Relates (Cross-References)

Depends On

• skill://lev-research - Provides recon before CDO execution
• skill://lev-find - Context gathering (legacy skill name for lev get backend; prior art, existing patterns)
• skill://bd - Epic tracking for multi-session workflows

Works With

• skill://planning - CDO outputs (proposals) feed into spec authoring
• skill://work - Alignment and lifecycle gates run in work
• skill://ralph - Adversarial validation for debug workflows

Blocks/Enables

• Blocks: Execution until DoR validation passes
• Enables: Multi-perspective analysis without groupthink

Lifecycle Position

• Before: Raw problem statement or user query
• After: FINAL-synthesis.md → proposal.md or report.md
• Parallel: Can run multiple CDO workflows simultaneously (different problems)

---

Sub-Skills Reference

CDO is organized into 4 sub-skills for progressive disclosure:

1. router/SKILL.md - Classification & Routing

Content:

• Classification logic (intent, confidence, complexity)
• DoR validation gates (4 gates)
• Routing decision table
• Message footer protocol

Use when: Understanding how CDO routes work

---

2. workflows/SKILL.md - Execution Patterns

Content:

• Quick/Base/Deep/Epic workflow examples
• Convergence criteria (critical for avoiding infinite loops)
• Resonance loops
• Agent dispatch patterns
• BD integration for Epic workflows

Use when: Building or executing CDO workflows

---

3. debug/SKILL.md - RCA Workflow

Content:

• 7-step debug template (REPRODUCE → VALIDATE)
• Integration with lev-ultra router
• Minimal fix anti-scope-creep patterns

Use when: Debugging complex issues systematically

---

4. skill-discovery/SKILL.md - lev-catalog Integration

Content:

• Semantic search for skills (axioms, hidden-gems)
• Tag-based browsing
• Discovery-first workflow construction
• 40+ skill catalog reference

Use when: Building Deep/Epic workflows from scratch

---

Quick Start

Step 1: Classification

lev cdo classify "Should we migrate to microservices?"

Output:

intent: question
confidence: 0.75
complexity: base
route_to: workflows/base

---

Step 2: Execution

lev cdo "Should we migrate to microservices?"

Process:

1. Router classifies (complexity: base)
2. Validates DoR (prior art check, architecture compliance)
3. Dispatches Base workflow:
   • Turn 1: Parallel (argue-for, argue-against)
   • Turn 2: Synthesis
4. Outputs: FINAL-analysis.md

---

Step 3: Review Output

cat tmp/microservices-analysis-20260128/FINAL-analysis.md

Contains:

• Pros (from Turn 1a)
• Cons (from Turn 1b)
• Synthesis with recommendation

---

Anti-Patterns to Avoid

Don't	Do Instead
Use CDO for simple questions	Direct answer or Quick workflow
Skip DoR validation	Always validate before Deep/Epic
Let orchestrator synthesize	Dispatch separate synthesis agent
Build workflows from scratch	Use skill-discovery to find existing patterns
Run Deep without convergence criteria	Define exit conditions upfront

---

Related Documentation

Architecture:

• BD Issue: lev-001b (lev-core Unified Orchestration)
• Protocol: skill:// (lev-2598), lev:// (lev-pno8)
• Proof of Concept: ~/lev/workshop/poc/skills/tmp/axiom-*/

Integration:

• lev-ultra router: Routes ultra-mode keywords (uld, ult, ulr, ulw)
• BD tracking: Epic workflows create BD structure
• FlowMind: YAML-first compilation (not runtime consumer)

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 lev-cdo, 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.