Evie Dev Framework

A template repository for building software with Claude Code. Slash commands produce specification artifacts and then build against them, scaling from small bounded changes up to full feature builds.

This repo is a template

Clone or fork it; the clone becomes the app you're building. Several files ship with placeholders that get filled in as you use the template:

• This README.md — describes the framework; replace it with one for your app.
• CLAUDE.md — # [Project Name] and the run/test/deploy sections.
• declaration.md — what / why / for whom / out of scope / Shape / Roadmap.
• constitution.md's ## Testing section — framework and run command, populated on first test generation.
• features/ — empty; folders are created per feature as commands run.

The template itself has no executable code and no tests; the quality gates in constitution.md apply to copied apps, not to this template.

Getting started

Open the cloned repo in Claude Code and:

1. /setup — fills CLAUDE.md (name, description, run/test, deploy target) and writes a README stub.
2. /declaration — populates declaration.md (intent + Shape + Roadmap). Runs automatically as the second phase of /setup on a fresh clone.
3. Pick your first piece of work and run its command.

Commands at a glance

Command	Use it for
/setup	One-time fill of CLAUDE.md + README stub on a fresh clone
/declaration	Write or refine the project declaration (intent, Shape, Roadmap)
/patch	A bounded change — bug, tweak, small feature — in one session
/feature	Declare a feature before speccing it
/spec	Drive requirements → architecture → adversarial → tests → DAG automatically
/requirements, /architecture, /adversarial, /tests, /dag	The spec steps, run manually for closer involvement
/build	Execute the DAG wave-by-wave and open a PR
/next	Run one build wave (resume a partial DAG)
/retro	Capture what was learned after a feature build
/upgrade	Pull the latest framework files into this project

See user-guide.md for how the requirements ↔ architecture loop converges, how adversarial findings flow back, and how the DAG-driven build works.

Development environment

Designed for Claude Code in the cloud sandbox attached to this GitHub repo. The sandbox provisions a working branch per session; commands commit to it and open PRs against main when work is complete. Local Claude Code works too — the only sandbox-specific accommodation is that user-global preferences live inside CLAUDE.md rather than ~/.claude/CLAUDE.md (which doesn't exist in the sandbox).