Help

---
allowed-tools:
  - Read
description: "Usage: /iris:help - Display Iris framework overview and commands"
---

**OUTPUT THE FOLLOWING EXACTLY AS IS - DO NOT ADD YOUR OWN TEXT**:

● ██ ██████ ██ ███████ ██ ██ ██ ██ ██ ██ ██████ ██ ███████ ██ ██ ██ ██ ██ ██ ██ ██ ██ ███████

       Autonomous Development Engine
       ------------------------------

IRIS transforms ideas into working applications autonomously. Provide requirements, enable dangerous permissions, and watch IRIS build your project from planning to completion without human intervention.

🚀 PRIMARY COMMAND

/iris:autopilot - 🚀 AUTONOMOUS DEVELOPMENT (Start here!)

🛠️ Manual Commands (for debugging/control)

/iris:plan - Manual sprint planning /iris:execute [task-id] - Manual task execution /iris:validate [milestone] - Manual milestone validation /iris:refine - Ralph-style iterative refinement /iris:document [flags] - Generate/update documentation /iris:audit [scope] - Essential security analysis

⚡ Quick Start - Autopilot Mode

1. Enable dangerous permissions: export CLAUDE_DANGEROUS_MODE=true
2. Run autopilot: /iris:autopilot "Build a task management API"
3. Monitor progress: watch cat PROJECT_STATUS.md
4. Check logs: tail -f .tasks/autopilot.log

🔧 Manual Mode (for step-by-step control)

1. Plan: /iris:plan "Your requirements"
2. Execute: /iris:execute
3. Validate: /iris:validate
4. Refine: /iris:refine (Ralph-style improvement loop)
5. Document: /iris:document
6. Audit: /iris:audit

📚 Document Command Flags

--standalone Analyze existing project without IRIS database --research Generate TECH_DECISIONS.md from research results --milestone M1 Update docs for specific milestone --final Generate COMPLETION_REPORT.md with KPIs

🔬 How Research Works

IRIS uses dynamic, PRD-driven research with prose-orchestration:

1. Foundation - Analyze PRD, select research opportunities
2. Parallel - Launch subagents to research technologies
3. Reconcile - Verify coherence, commit approved stack

Research output: TECH_DECISIONS.md (technology choices with rationale)

🔄 How Refinement Works (Ralph Wiggum Philosophy)

After development, IRIS runs iterative refinement loops:

1. Review - Parallel subagents analyze code with fresh context
2. Aggregate - Combine findings, prioritize by severity
3. Refine - Single subagent improves code toward PRD intent
4. Validate - Run tests (backpressure, not a gate)
5. Repeat - Fixed iterations (5-10 based on complexity)

Key principles:

• Fresh context each iteration (no accumulated baggage)
• Progress persists in files/git, not context
• "Improve, not just fix" - enhance toward PRD intent
• Never exit early - run all iterations

IRIS uses adaptive complexity scaling, prose-orchestration, dynamic research, and Ralph Wiggum refinement to deliver quality prototypes autonomously.