War Room: AI Decision Framework with Devil's Advocate

Purpose and Significance

War Room is an open-source multi-agent decision framework that applies structured protocols to software architecture, product strategy, and critical business decisions. Unlike standard multi-agent systems that coordinate execution, War Room enforces a mandatory "devil's advocate" workflow—the CHAOS agent systematically attacks assumptions, surfaces root causes, and produces counter-proposals that competing approaches miss. In a documented case study, the same project briefing produced 10 over-scoped features under standard review but only 8 justified features through War Room, with 6 features eliminated entirely (saving 5 developer days) and risks elevated from surface-level lists to root-cause analysis with documented switch costs.

Key Features

• CHAOS Agent (Built-in Devil's Advocate) — Shadows every decision wave, rates outcomes as SURVIVES, WOUNDED, or KILLED, and generates counter-proposals the primary agents overlook.
• 19 Structured Decision Protocols — Mandatory constraints across four decision pillars (Essential 7 + Advanced 12), including Test State, Five Whys, Ignorance Declaration, Via Negativa, Plan B, and Pre-Mortem frameworks.
• Wave-Based Agent Orchestration — Specialists execute in dependency order (Foundation → Specialists → Builders → Validators), not in parallel, ensuring decisions cascade correctly.
• Domain-Specific Agent Roles — 16+ specialist personas: ARCH, PM, DEV, SEC, UX, QA, MKT, RESEARCH, FINANCE, LEGAL, plus custom roles (AI-ENG, AUDIO, DATA, OPS) for problem-specific needs.
• Append-Only Decision Log — Immutable DECISIONS.md tracks all choices, rationale, and evolution; full audit trail for post-mortems and future reference.
• Consolidated Blueprint Output — Produces PRD, architecture specs, risk analysis, lessons learned, and inter-agent communication in structured artifacts.
• MIT Licensed, Framework-Agnostic — Runs standalone or integrates with CrewAI, AutoGen, or MetaGPT; shell-based initialization, no vendor lock-in.

How It Works

Initialization and Setup

Start with three commands: initialize a war room with scripts/init_war_room.sh, write your project brief (constraints, scope, goals), and inject the DNA template containing all 19 protocols. The system then auto-spawns agents based on problem complexity—4 to 13 specialists tailored to your brief.

Wave Protocol Execution

Agents run in four dependency waves. Wave 1 (Foundation) locks architecture, security, and product decisions that others depend on. Wave 2 (Specialists) builds on Wave 1 with UX, audio, and AI engineering decisions. Wave 3 (Builders) implements based on prior waves. Wave 4 (Validators) stress-tests everything with QA, marketing, and CHAOS review. The CHAOS agent shadows every wave—not just the final output—catching failures early.

Decision Protocols in Practice

Each decision must survive seven essential protocols: Test State (argue the opposite), Five Whys (dig to root cause), Ignorance Declaration (mark unknowns explicitly), Via Negativa (list what to remove first), Plan B (backup with switch cost), Pre-Mortem (production failure scenarios), and CHAOS Review (adversarial stress test). Advanced power users unlock 12 additional protocols: Dialectic, Mirror Test, Ripple Analysis, Tension Map, Causal Chain Verification, Tempo Tagging, Create-Then-Constrain, Barbell Strategy, and Lessons Permanent.

CHAOS Agent Methodology

CHAOS rates every decision on three outcomes: SURVIVES (withstands scrutiny), WOUNDED (valid but has weaknesses), or KILLED (doesn't hold up, needs rethinking). It also produces counter-proposals—alternative approaches bypassed by primary agents. In the published test case, CHAOS identified that 4 of the top 5 failure scenarios stemmed from a single dependency, a critical finding that 12 other specialists had missed.

Getting Started

For OpenClaw users: Run openclaw skill install war-room to add War Room as a verified skill.

Manual setup: Clone the repository, run bash scripts/init_war_room.sh my-project, edit war-rooms/my-project/BRIEF.md with your scope and constraints, inject the DNA template, then invoke your agent with the instruction: "Run a war room on my-project." The agent handles orchestration, spawning, and wave execution automatically.

Output location: Results appear in war-rooms/my-project/ with structured subdirectories: agent outputs, inter-agent communications, decision logs, blockers, and a consolidated BLUEPRINT.md artifact.

Who It's For

• Product and Engineering Teams — Use War Room to validate architecture choices, scope product roadmaps, and surface hidden risks before commitment.
• Startup Founders — Stress-test pivot decisions, market positioning, and business model assumptions with systematic devil's advocacy.
• Open-Source Maintainers — Plan launch strategy, competitive positioning, and distribution channels through multi-stakeholder analysis.
• Enterprise Decision-Makers — When a decision costs weeks of work if wrong, War Room produces auditable decision trails with explicit unknowns and counter-proposals.
• AI and LLM Teams — Teams building with multi-agent systems who want structured decision protocols beyond basic coordination frameworks.

When to Use War Room

Ideal Use Cases

• Decisions with high execution cost (weeks of work, significant resource commitment)
• Complex trade-offs across multiple domains (architecture, product, business, compliance)
• Situations requiring multiple perspectives but lacking multiple subject-matter experts
• Pre-commitment stress testing of existing plans or strategies
• Situations where "ship first, learn later" carries unacceptable risk

When Not to Use

• Simple, well-defined tasks with clear optimal solutions
• Situations requiring immediate answers over deep analysis
• Decisions already made that only need execution planning

Differentiation from Existing Frameworks

CrewAI, AutoGen, and MetaGPT are agent orchestration frameworks—they help you run multiple agents in sequence or parallel. War Room is a decision methodology that uses multiple agents as the vehicle. The core difference is the DNA: 19 mandatory protocols that force agents to question assumptions, declare ignorance, attack their own conclusions, and produce counter-proposals. War Room can run on top of these frameworks if desired, but the protocols are what distinguish it.

Real-World Examples

Software Example: "Build a macOS app for AI music generation" → 4 waves, 13 agents, 57 decisions, 32 documents, complete implementation blueprint in 35 minutes, with documented failure scenarios and mitigation strategies.

Business Example: "Should we pivot from B2C to B2B?" → CHAOS attacks both models, Five Whys reveals the root cause isn't the model itself, counter-proposal surfaces a third viable option.

Launch Example: "Plan the launch strategy for an open-source project" → MKT agent develops positioning, RESEARCH conducts competitive analysis, CHAOS identifies a gap in the distribution strategy.

Contributing and Community

War Room is open source to enable knowledge sharing. Contribution areas include: new domain-specific agent roles, refinements to the 19 protocols, integrations with additional orchestration frameworks, and shared lessons from production war rooms. See SKILL.md and agent-roles.md in the repository for templates and contribution guidelines.

Resources

• Repository: github.com/maxkle1nz/war-room
• Quick Start Guide: Included in README; initialization takes under 5 minutes
• DNA Template: Complete protocol definitions in references/dna-template.md
• Agent Roles: Full roster and custom role template in agent-roles.md
• Wave Protocol: Execution details in wave-protocol.md
• OpenClaw Integration: openclaw skill install war-room for managed deployment

License: MIT. Free to use, modify, and redistribute.