Claude Auto-Dev

Autonomous development workflow for Claude Code. Say what you want to build — Claude handles the rest.

Quick Start

# Mac/Linux
git clone https://github.com/djnsty23/claude-auto-dev ~/claude-auto-dev
cd ~/claude-auto-dev && ./install.sh

# Windows (PowerShell)
git clone https://github.com/djnsty23/claude-auto-dev $env:USERPROFILE\claude-auto-dev
cd $env:USERPROFILE\claude-auto-dev; .\install.ps1

Then run claude and say brainstorm — Claude scans the codebase, proposes improvements, creates stories, and works through them autonomously.

Full Environment Restore (post-reinstall / new machine)

Use this section when: fresh OS install, new dev machine, or onboarding to an existing claude-auto-dev setup.

For humans (manual)

# Windows — run in an admin PowerShell
# 1. Prereqs
winget install -e Git.Git OpenJS.NodeJS GitHub.cli Doppler.doppler Microsoft.VisualStudioCode

# 2. claude-auto-dev itself (this repo)
git clone https://github.com/djnsty23/claude-auto-dev $env:USERPROFILE\claude-auto-dev
cd $env:USERPROFILE\claude-auto-dev; .\install.ps1 -Full

# 3. Authenticate all services
gh auth login              # GitHub — browser flow
doppler login              # Doppler — browser flow
# claude login             # If Claude Code CLI installed separately

# 4. Restore personal memory + config (PRIVATE repo — requires gh auth from step 3)
#    This assumes you have a `claude-memory` private repo (see memory-backup skill)
gh repo clone <your-username>/claude-memory $env:USERPROFILE\claude-memory
cd $env:USERPROFILE\claude-memory
# Follow RECOVERY.md in that repo for the per-user restore

# Mac/Linux equivalent
brew install git node gh dopplerhq/cli/doppler
git clone https://github.com/djnsty23/claude-auto-dev ~/claude-auto-dev
cd ~/claude-auto-dev && ./install.sh --full
gh auth login
doppler login
gh repo clone <your-username>/claude-memory ~/claude-memory
cat ~/claude-memory/RECOVERY.md

For Claude (you're reading this as the agent)

If the user points you at this repo post-reinstall and says "configure everything", walk this order — stop and ask for authorization at each interactive step:

Detect OS + missing prereqs — check for git, node, gh, doppler. Install any missing via the OS package manager (winget on Windows, brew on macOS, distro-appropriate on Linux).
Verify ~/.claude/ install — if this repo is cloned to ~/claude-auto-dev, run install.ps1 -Full (Windows) or install.sh --full (Unix).
Check gh auth status — if broken, tell the user to run gh auth login in their terminal. DO NOT try to run the OAuth flow yourself (requires browser).
Check doppler me — if broken, tell the user to run doppler login. Same reason.
Offer memory restore — ask the user if they have a private claude-memory repo. If yes, gh repo clone <user>/claude-memory ~/claude-memory, then copy memory/projects/* into ~/.claude/projects/*/memory/, copy CLAUDE.md to ~/.claude/CLAUDE.md, copy rules/ to ~/.claude/rules/. Set ~/.claude/.memory-backup-path to the restored path so future memory backup now calls work.
Per-project doppler setup — for each repo under ~/Downloads/code/ or ~/code/ that has a doppler.yaml, ensure the bind is live: cd <repo> && doppler run -- node -e "console.log(Object.keys(process.env).length)" should return a sensible number. If not, doppler setup manually.
Report — produce a checklist showing what's restored, what the user still needs to touch (e.g. rotating a stale token, reconnecting OneDrive).

Never:

Autonomously run gh auth login or doppler login — they open browsers and block waiting on user input.
Commit to any repo without the user having configured git config user.email / user.name.
Restore secrets into local .env files — they belong in Doppler only.

You: brainstorm         → Claude finds 5 improvements
You: brainstorm apply   → Creates 5 stories in prd.json
You: auto               → Works through all tasks
You: ship               → Review, security scan, deploy

Install

Prerequisites: Git and Node.js 18+ (the installer adds Claude Code for you if it's missing).

Options:

Flag	What it does
`--full` / `-Full`	Also install rules + settings templates
`--init` / `-Init`	Scaffold `prd.json` in the current project
`--force`	Back up and overwrite files that collide with shipped names

What it does:

Copies skills/, hooks/, and agents/ into ~/.claude/
Writes ~/.claude/.auto-dev-installed.json — a sidecar recording exactly what got installed (used by uninstall)
Adds an update-dev command to your shell
Saves your repo path so updates are portable

Collision handling. Install leaves user-owned files alone. If ~/.claude/skills/ or ~/.claude/hooks/ already contains something with the same name as a shipped file (and it's not byte-identical), install refuses and lists the conflicts. Re-run with --force to back them up to ~/.claude/.user-backup-<timestamp>/ before overwriting.

Your own skills with different names (e.g. my-company-skill/) are always preserved.

Updates

You: update dev
Claude: [pulls latest, re-syncs skills/hooks/agents] Updated to v7.5

Or from the shell:

update-dev

Commands

Say	Does
`brainstorm`	Scan codebase + live site, propose improvements
`brainstorm apply`	Create stories from the last brainstorm
`auto`	Work through all pending stories autonomously
`iterate`	Convergence loop: brainstorm → fix → re-scan until clean
`audit`	7-agent parallel quality audit
`review`	Code quality check (add `quick` or `deep`)
`ship`	Build, test, review, deploy, verify
`scan` / `qa`	Live site QA (visual + a11y + console)
`fix`	Debug issues
`commit`	Conventional commit + push + PR
`test`	Unit + browser tests
`security`	Pre-deploy security scan
`perf`	Core Web Vitals audit
`a11y`	WCAG 2.1 AA audit
`design` / `ui`	UI design with anti-slop checklist
`progress`	Show sprint progress
`sprint`	Create/advance sprint
`update dev`	Sync latest from GitHub

See skills/commands.md for the full list.

Quick fixes — skip the ceremony. For small tasks, just describe what you want. No auto, no sprints, no prd.json:

fix the button overflow on mobile
add loading state to the dashboard

Workflow

brainstorm → scans codebase + live site, proposes improvements
auto       → implements all pending stories + visual verification
ship       → review + security + deploy + post-deploy scan
iterate    → brainstorm → fix → re-scan loop until clean

Visual verification. Start your dev server before auto. Claude screenshots pages (desktop + mobile) after each UI change and catches console errors before marking tasks complete.

Image auto-scan. Attach a screenshot to any turn and Claude surfaces every distinct issue it sees, not only the one you asked about. Add [focus] in your message to opt out.

Files

Global (~/.claude/):

skills/                     # 36 skills (copied from repo)
hooks/                      # 6 hooks
agents/                     # 4 specialized agents
rules/                      # Workflow/security/design rules (only with --full)
settings.json               # Merged with your existing settings (only with --full)
repo-path.txt               # Points to your clone location
.auto-dev-installed.json    # Install sidecar — what this install put on disk

Per project:

prd.json       # Tasks and sprint history

Tasks use passes: null (pending), true (done), or "deferred".

Uninstall

# Mac/Linux
cd ~/claude-auto-dev && ./uninstall.sh

# Windows (PowerShell)
cd $env:USERPROFILE\claude-auto-dev; .\uninstall.ps1

# Preview first (no changes)
./uninstall.sh --dry-run        # or: .\uninstall.ps1 -DryRun

Uninstall reads .auto-dev-installed.json and removes exactly what this install created. Your own skills, hooks, agents, and user-modified rules stay. It strips auto-dev hook entries from ~/.claude/settings.json without touching other entries, then deletes the sidecar and repo-path.txt.

Manual step: remove the update-dev function from your shell profile (~/.bashrc, ~/.zshrc, or the PowerShell profile) if you don't want it anymore.

Troubleshooting

Install refuses with "REFUSING TO OVERWRITE". You have a file or directory with the same name as something we ship, and it's not byte-identical to ours. Either rename yours or re-run with --force (which backs yours up first).

Not seeing updates? Check ~/.claude/repo-path.txt points to your clone. Manual fallback: cd ~/claude-auto-dev && git pull && update-dev.

Skills not loading? Verify ~/.claude/skills/ exists and contains skill directories. Restart your terminal after install.

Hook errors? Requires Node.js 18+. Check with node -v. Hooks fail silently by design — run node validate.js to check consistency.

License

MIT

Name	quality
Description	Enforces type safety, design tokens, and all UI states.

quality

SKILL.md