Draft

Draft - Context-Driven Development

Draft is a methodology for structured software development: Context → Spec & Plan → Implement

Red Flags - STOP if you're:

• Jumping straight to implementation without reading existing Draft context
• Suggesting /draft:implement before a track has an approved spec and plan
• Not checking draft/tracks.md for existing active tracks before creating new ones
• Skipping the recommended command and going freeform
• Ignoring existing .ai-context.md, product.md, tech-stack.md, or workflow.md context

Read context first. Follow the workflow.

Two-Tier Command Architecture

Primary Workflow (4 commands)

init → new-track → implement → review
                       ↑           |
                       └───────────┘  (auto-invoked at phase boundaries)

Specialist Commands (21 commands)

Setup & Navigation:

Planning & Architecture:

Code Quality:

Debugging:

Operations:

Authoring:

Integration:

Quick Start

1. First time? Run /draft:init to initialize your project
2. Starting a feature? Run /draft:new-track "your feature description"
3. Ready to code? Run /draft:implement to execute tasks
4. Check progress? Run /draft:status

Core Workflow

Every feature follows this lifecycle:

1. Setup - Initialize project context (once per project)
2. New Track - Create specification and plan
3. Implement - Execute tasks with TDD workflow
4. Verify - Confirm acceptance criteria met
5. Quality - Run quality commands (see guide below)

Auto-invocations: The primary workflow has built-in quality gates — /draft:implement auto-invokes /draft:review at phase boundaries, and /draft:review auto-invokes /draft:coverage when TDD is enabled.

Quality Commands — When to Use Which

Four commands form an audit spectrum from quick to narrow to broad to deep:

Decision Guide

• Quick sanity check? → /draft:quick-review — fast 4-dimension review, no track context needed
• Just finished a track? → /draft:review — validates against spec, checks quality gates
• Suspicious of bugs across the codebase? → /draft:bughunt — 14-dimension sweep with verification protocol
• Shipping a module to production? → /draft:deep-review — ACID compliance, resilience, observability audit
• Want everything? → /draft:review full (includes bughunt), then /draft:deep-review for critical modules

Relationship to Built-in Bug Hunt Agents

Some AI tools provide built-in bug hunt agents (e.g., Claude Code's bughunt agent). These are complementary to /draft:bughunt — the built-in agents offer fast parallel sweeps with auto-fix, while Draft's bughunt adds context-aware analysis using your architecture, tech-stack, and product context for better false-positive elimination. For maximum coverage, run both.

Context Files

When draft/ exists, these files guide development:

• draft/architecture.md - Source of truth: comprehensive human-readable engineering reference
• draft/.ai-context.md - Derived from architecture.md: token-optimized AI context (200-400 lines)
• draft/product.md - Product vision and goals
• draft/tech-stack.md - Technical constraints
• draft/workflow.md - TDD and commit preferences
• draft/guardrails.md - Hard guardrails, learned conventions, learned anti-patterns
• draft/tracks.md - Active work items

Status Markers

Used throughout plan.md files:

Intent Mapping

You can also use natural language:

Need Help?

• Run /draft (this command) for overview
• Run /draft:status to see current state
• Check draft/tracks/<track_id>/spec.md for requirements
• Check draft/tracks/<track_id>/plan.md for task details

Converted and distributed by TomeVault — claim your Tome and manage your conversions.