Autonomous Engineering Workflow System

A structured development workflow system for Claude Code that integrates practices from elite organizations into a cohesive, learning-enabled development process.

Philosophy

This system draws from methodologies proven in high-stakes environments:

Source	Practice	Application
Navy SEALs	After Action Review (AAR)	Structured learning after each cycle
Toyota	Kanban, Andon Cord	Flow management, stop-the-line quality
High Reliability Organizations	5 HRO Principles	Failure awareness, operational sensitivity
Bridgewater	Believability Weighting	Track prediction accuracy over time
Pixar	Braintrust	Candid feedback without authority
Military	Commander's Intent	Clear purpose enables autonomous decisions
Gary Klein	Pre-mortem	Prospective hindsight for risk identification
Netflix	Chaos Engineering	Controlled experiments to find weaknesses
Emergency Medicine	Triage	Systematic prioritization under pressure

Quick Start

# Clone to your Claude Code config directory
git clone <repo> ~/.claude

# Start a development cycle
/dev-cycle start add user authentication

# Or use individual commands
/plan add payment processing
/execute-plan docs/plans/20251201-143022-payment-processing.md
/critical-review docs/plans/20251201-143022-payment-processing.md

Core Workflow

flowchart TD
    Start["/dev-cycle start"]

    Start --> Triage["/triage<br/>(what?)"]
    Start --> Plan["/plan<br/>(how?)"]
    Start --> Spike["/spike<br/>(uncertain)"]

    Triage --> Plan
    Spike --> Plan

    Plan --> PreMortem["/pre-mortem<br/>(risks?)"]
    Plan --> Ensemble["/ensemble<br/>(feedback?)"]

    PreMortem --> Execute
    Ensemble --> Execute

    subgraph Execute["/execute-plan"]
        Demo["Demo → Review<br/>FAITHFUL / DEVIATE / DIVERT"]
    end

    Execute --> Handoff["/handoff quick"]
    Execute --> Review["/critical-review"]
    Execute --> RedTeam["/red-team<br/>(security)"]

    Review --> Pass{PASS}
    Review --> NeedsWork{NEEDS_WORK}

    Pass --> AAR["/aar<br/>(learn)"]
    NeedsWork --> FailureAnalysis["/failure-analysis"]

    AAR --> Complete([COMPLETE])
    FailureAnalysis --> Complete

    Complete -.->|periodically| Retro["/retro<br/>(improve)"]

Loading

Architecture

~/.claude/
├── commands/               # Slash commands (*.md)
│   ├── plan.md            # /plan - Structured planning with COA scoring
│   ├── execute-plan.md    # /execute-plan - Demo/review execution pattern
│   ├── critical-review.md # /critical-review - Quality evaluation
│   ├── dev-cycle.md       # /dev-cycle - Workflow orchestration
│   ├── aar.md             # /aar - After Action Review
│   └── ...                # 20+ additional commands
│
├── skills/                 # Auto-activated capabilities
│   ├── hro-principles/    # High Reliability Organization mindset
│   ├── coa-analysis/      # Course of Action scoring
│   ├── failure-analysis/  # Root cause analysis (5 Whys)
│   ├── gap-analysis/      # Coverage gap identification
│   ├── dynamics-analysis/ # Boundary-first system exploration
│   └── ...                # 10+ additional skills
│
├── global-knowledge/       # Cross-project learnings
│   ├── patterns.md        # Proven approaches
│   ├── anti-patterns.md   # Failure modes to avoid
│   ├── lessons.md         # Key insights
│   ├── risk-catalog.md    # Known risk patterns
│   └── insights.md        # Meta-learnings about process
│
├── templates/              # Project initialization templates
│   ├── Makefile.claude-hooks.example
│   ├── plan-template.md
│   ├── adr-template.md
│   └── onboarding.md
│
└── settings.json           # Hooks configuration

Commands Reference

Orchestration

Command	Purpose
/dev-cycle start <goal>	Begin full development cycle
/dev-cycle start --lite <goal>	Lightweight cycle for simple tasks
/dev-cycle resume	Resume interrupted cycle
/dev-cycle status	Show current cycle state

Planning & Prioritization

Command	Purpose
/plan <goal>	Structured planning with COA scoring
/triage <items>	Prioritize competing work items
/spike <question>	Time-boxed uncertainty reduction
/adr new <title>	Record architecture decision

Risk & Preparation

Command	Purpose
/pre-mortem <plan>	Identify risks before starting
/red-team <target>	Adversarial security review
/ensemble <problem>	Multi-perspective analysis

Execution

Command	Purpose
/execute-plan <plan>	Run plan with demo/review pairs
/handoff quick	Quick progress checkpoint
/runbook run <name>	Execute codified procedure

Review & Analysis

Command	Purpose
/critical-review <plan>	Full quality evaluation
/failure-analysis [target]	Root cause analysis
/gap-analysis [focus]	Find coverage gaps
/dynamics <target>	Systematic parameter exploration

Learning & Improvement

Command	Purpose
/aar <plan>	After Action Review
/retro [range]	Meta-retrospective across cycles
/knowledge query <terms>	Search knowledge base
/knowledge promote <ID>	Promote to global knowledge

Skills (Auto-Activated)

Skills activate automatically based on context:

Skill	Activates When
hro-principles	Throughout all work (5 HRO principles)
coa-analysis	Decision points with multiple options
failure-analysis	Errors, test failures, exceptions
gap-analysis	Assessing completeness or coverage
dynamics-analysis	Analyzing system behavior or limits
cynefin	Problem classification (Clear/Complicated/Complex/Chaotic)
commander-intent	Execution decisions requiring judgment
incident-command	Production issues requiring response

Demo/Review Pattern

During /execute-plan, each task spawns paired agents:

flowchart LR
    Impl["IMPLEMENTATION<br/>AGENT (Demo)"] --> Review["REVIEW<br/>AGENT"]

    Review --> Faithful["✓ FAITHFUL"]
    Review --> Deviation["✗ DEVIATION"]
    Review --> Diversion["↪ DIVERSION"]

Loading

Result	Meaning	Action
FAITHFUL	Matches plan exactly	Proceed
DEVIATION	Differs without reason	Reject, redo
DIVERSION	Discovered limitation	Document, proceed

Knowledge Base

Two-Tier Architecture

~/.claude/global-knowledge/     ◀── Cross-project (always available)
     │
     ├── patterns.md            PATTERN-XXX: Universal patterns
     ├── anti-patterns.md       ANTI-PATTERN-XXX: Universal failures
     ├── lessons.md             LESSON-XXX: Universal lessons
     └── risk-catalog.md        RISK-CAT-XXX: Universal risks

<repo>/docs/knowledge/          ◀── Project-specific
     │
     ├── patterns.md            Project patterns
     ├── anti-patterns.md       Project failure modes
     ├── lessons.md             Project lessons
     └── metrics.md             COA prediction tracking

Knowledge Flow

flowchart LR
    Work["Project Work"] --> AAR["/aar captures learnings"]
    AAR --> Local["docs/knowledge/"]
    Local --> Promote["/knowledge promote"]
    Promote --> Global["~/.claude/global-knowledge/"]

Loading

COA Scoring

When evaluating options, use weighted scoring:

Criterion	Weight	5 = Best
Impact	×2	Fully solves problem
Effort	×1	Minimal effort
Risk	×1.5	Low risk
Dependencies	×1	Self-contained
Reversibility	×1	Easily reversible

Max Score: 32.5 | Strong: 26+ | Good: 20-25 | Acceptable: 14-19

Makefile Integration

Copy templates/Makefile.claude-hooks.example to your project and customize:

# Hook targets (called automatically)
check:           # After agent tasks - quick verification
lint-changed:    # After file edits - style feedback
validate:        # End of response - comprehensive check

# Analysis targets (called by commands)
test-report:     # Detailed output for /failure-analysis
coverage:        # Coverage report for /gap-analysis
security-scan:   # Security findings for /red-team

Hooks are configured in settings.json:

• check failures trigger the failure-analysis skill
• lint-changed provides immediate feedback on edits
• validate runs comprehensive checks before human review

Installation

1. Clone to Claude Code config directory:

   git clone <repo> ~/.claude

2. For each project, initialize the knowledge base:

   /dev-cycle start <your first goal>
   # Follow prompts to initialize docs/knowledge/ from templates

3. Optional: Add Makefile hooks to your project:

   cp ~/.claude/templates/Makefile.claude-hooks.example ./Makefile
   # Customize targets for your stack

Entry ID Conventions

Type	Format	Example
Pattern	PATTERN-XXX	PATTERN-003
Anti-Pattern	ANTI-PATTERN-XXX	ANTI-PATTERN-012
Lesson	LESSON-XXX	LESSON-007
Risk (catalog)	RISK-CAT-XXX	RISK-CAT-005
ADR	ADR-XXX	ADR-015
Debt	DEBT-XXX	DEBT-008
Incident	INC-YYYYMMDD-HHMMSS	INC-20251128-143022

License

AGPL-3.0 - See LICENSE for details.