Home

Dissonance Labs Starter

Table of Contents

• Mission & Philosophy
• Two-Lane Development Model
• Guardrails
• Glossary
• Quick Links

Mission & Philosophy

Optimize for: Fast, correct TDD; secure defaults; zero-drift docs.

This starter implements AI-assisted development with dual lanes for different complexity levels, ensuring quality while maintaining velocity.

Two-Lane Development Model

Planning Lane vs Execution Lane

Simple Workflow (Execution Lane)

• Small changes, bug fixes, refactoring
• Single slash command: /dev:plan-feature
• Direct implementation with quality gates

Spec-Driven Workflow (Planning Lane)

• Complex features, new business logic, architectural decisions
• Multi-step: /specify → /plan → /tasks → implement
• Structured planning with GPT-5 collaboration

When to Use Which

Use Simple (/dev:plan-feature) for:

• Bug fixes and small improvements
• Single-component changes
• Documentation updates
• Quick prototypes

Use Spec-Driven (/specify) for:

• New features with multiple components
• Complex user workflows
• Anything requiring architectural decisions
• When collaborating with GPT-5 on planning

Guardrails

Branch Isolation

• All bot work happens on bots/claude/* branches
• Human-only triggers (bots cannot activate bots)
• Path restrictions: no access to .github/workflows/**, .env*

Promote Label System

• Bot PRs require maintainer-added promote label before merge
• Advisory-first approach with measurable false-positive tracking
• Human review remains authoritative

AI Lane Control

• Instant disable: Set CLAUDE_ENABLED=false to disable AI-powered PR reviews
• Security disable: Set CLAUDE_SECURITY_ENABLED=false to disable AI security scanning
• Emergency stop: Both flags provide immediate control over AI automation

Quality Gates

• TDD first: tests before implementation
• Conventional commits only
• Least-deps: justify new packages
• No secrets in prompts/code

Glossary

Kernel: Specification document (Full or Micro template) that defines requirements and acceptance criteria

Doctor: Quality pipeline script (npm run doctor) that checks project health, links, and compliance

Slash Command: Structured workflow command (e.g., /dev:plan-feature, /review:ai-powered) that encapsulates multi-step processes

Bot Branch: Isolated branch (bots/claude/*) where AI agents perform work safely

Advisory Review: AI-powered feedback that provides suggestions but doesn't block merges

Promote Label: Manual label (promote) required on bot PRs before merge, ensuring human oversight

Quick Links

• Spec System Templates & Usage
• Complete Commands Reference
• Quality Gates & Pipeline
• Architecture & Stack Overview
• Getting Started Guide
• AI Collaboration Best Practices

---

This wiki serves as LLM context and human reference. For operational commands and daily workflow, see CLAUDE.md in the repository.