npx projecta-rrr

Why I Built This

I'm a solo developer. I don't write code — Claude Code does.

Other spec-driven development tools exist; BMAD, Speckit... But they all seem to make things way more complicated than they need to be (sprint ceremonies, story points, stakeholder syncs, retrospectives, Jira workflows) or lack real big picture understanding of what you're building. I'm not a 50-person software company. I don't want to play enterprise theater. I'm just a creative person trying to build great things that work.

So I built RRR. The complexity is in the system, not in your workflow. Behind the scenes: context engineering, XML prompt formatting, subagent orchestration, state management. What you see: a few commands that just work.

The system gives Claude everything it needs to do the work and verify it. I trust the workflow. It just does a good job.

That's what this is. No enterprise roleplay bullshit. Just an incredibly effective system for building cool stuff consistently using Claude Code.

— Projecta.ai

Vibecoding has a bad reputation. You describe what you want, AI generates code, and you get inconsistent garbage that falls apart at scale.

RRR fixes that. It's the context engineering layer that makes Claude Code reliable. Describe your idea, let the system extract everything it needs to know, and let Claude Code get to work.

Who This Is For

People who want to describe what they want and have it built correctly — without pretending they're running a 50-person engineering org.

Getting Started

npx projecta-rrr

That's it. Verify with /rrr:help inside your Claude Code interface.

Installed from inside Claude Code? Slash commands won't load until you restart:

1. Type exit to quit Claude Code
2. Run claude again
3. Then run /rrr:help

This resolves "Unknown skill: rrr:help" errors.

Staying Updated

RRR evolves fast. Check for updates periodically:

/rrr:whats-new

Update with:

npx projecta-rrr@latest

npx projecta-rrr --global   # Install to ~/.claude/
npx projecta-rrr --local    # Install to ./.claude/

git clone https://github.com/PA-Ai-Team/projecta-rrr.git
cd projecta-rrr
node bin/install.js --local

Recommended: Skip Permissions Mode

RRR is designed for frictionless automation. Run Claude Code with:

claude --dangerously-skip-permissions

[!TIP] This is how RRR is intended to be used — stopping to approve date and git commit 50 times defeats the purpose.

{
  "permissions": {
    "allow": [
      "Bash(date:*)",
      "Bash(echo:*)",
      "Bash(cat:*)",
      "Bash(ls:*)",
      "Bash(mkdir:*)",
      "Bash(wc:*)",
      "Bash(head:*)",
      "Bash(tail:*)",
      "Bash(sort:*)",
      "Bash(grep:*)",
      "Bash(tr:*)",
      "Bash(git add:*)",
      "Bash(git commit:*)",
      "Bash(git status:*)",
      "Bash(git log:*)",
      "Bash(git diff:*)",
      "Bash(git tag:*)"
    ]
  }
}

Projecta Preferred Pack

RRR includes a default stack optimized for rapid MVP development. Defaults are recommended but overrideable.

Where Defaults Live

Core Stack (always assumed)

• Framework: Next.js (App Router) + TypeScript
• Package Manager: npm
• UI: Tailwind CSS + shadcn/ui
• Testing: Vitest (unit) + Playwright (e2e)

Preferred Providers

Agent Stack (when agents needed)

• Orchestration: Mastra
• Agent Auth: Auth.dev
• Agent Mail: Agentmail
• Sandbox: E2B
• Browser Automation: Browserbase

Overrides

If you choose a non-default provider, RRR asks for a reason and records it in Deviation Notes:

## Deviation Notes

| Capability | Default | Chosen | Reason |
|------------|---------|--------|--------|
| auth | clerk | auth0 | Client requires Auth0 for SSO compliance |

Discouraged Providers

These are allowed but require explicit justification: Firebase, Supabase, Auth0, Vercel, PlanetScale.

MCP Auto-Setup

RRR includes an MCP registry that maps your selected providers to their MCP servers. After /rrr:new-project generates your MVP_FEATURES.yml:

npm run mcp:setup

This reads your feature selections and outputs the MCP configuration for Claude Code. Only MCPs you actually need get installed.

Always included: Context7 (docs), GitHub, Filesystem, Sequential Thinking.

Pushpa Mode (Autopilot)

Run phases overnight while you sleep. Pushpa Mode is an unattended runner that plans and executes phases sequentially, skipping any that require human verification.

bash scripts/pushpa-mode.sh
# or
npm run pushpa

What it does:

1. Preflights required API keys based on your MVP_FEATURES.yml
2. Iterates through phases in order
3. Plans any phase that doesn't have a plan yet
4. Executes phases automatically
5. Skips phases marked with HITL_REQUIRED: true (human verification needed)
6. Generates a morning report at .planning/PUSHPA_REPORT.md

Prerequisites:

• Run /rrr:new-project first (project must be initialized)
• Set all required API keys (script will check and warn)
• Recommend enabling YOLO mode in .planning/config.json

Where outputs live:

• Report: .planning/PUSHPA_REPORT.md
• Logs: .planning/logs/pushpa_*.log

HITL Convention: Plans that require human verification should include one of these markers:

• HITL_REQUIRED: true (canonical)
• HUMAN_VERIFICATION_REQUIRED
• MANUAL_VERIFICATION

Pushpa Mode will skip these phases and record them in the report for manual follow-up.

Where to run Pushpa Mode:

• Recommended: Run in a normal system terminal (outside Claude Code) for true unattended overnight runs:

  bash scripts/pushpa-mode.sh

• Running inside Claude Code works but can trigger approval prompts ("Do you want to proceed?").
• If the script detects Claude Code, it will prompt: Continue running Pushpa Mode inside Claude Code? (y/N) — default is No (recommended), so you can switch to an external terminal.

How It Works

1. Initialize Project (~10 minutes)

/rrr:new-project

One command, one flow. The system:

1. Questions — Asks until it understands your idea completely (goals, constraints, tech preferences, edge cases)
2. Research — Spawns parallel agents to investigate the domain (optional but recommended)
3. Requirements — Extracts what's v1, v2, and out of scope
4. Roadmap — Creates phases mapped to requirements

You approve the roadmap. Now you're ready to build.

Creates: PROJECT.md, REQUIREMENTS.md, ROADMAP.md, STATE.md, .planning/research/

2. Plan Phase

/rrr:discuss-phase 1   # Optional: clarify UI/UX/behavior decisions first
/rrr:plan-phase 1

discuss-phase (optional) — If the phase has gray areas (UI choices, UX flows, behavior decisions), discuss them first. Creates CONTEXT.md that guides planning. Skip if you trust the system's defaults.

plan-phase — The system:

1. Researches — Investigates how to implement this specific phase
2. Plans — Creates 2-3 atomic task plans with XML structure
3. Verifies — Checks plans against requirements, loops if needed

Ready when plans pass verification.

3. Execute Phase

/rrr:execute-phase 1

The system:

1. Runs plans in waves — Parallel where possible, sequential when dependent
2. Fresh context per plan — 200k tokens purely for implementation, zero degradation
3. Verifies code — Checks against phase goals when complete

4. Repeat

/rrr:plan-phase 2
/rrr:execute-phase 2
...
/rrr:complete-milestone   # When all phases done

Loop plan → execute until milestone complete. Ship your MVP. Start next milestone.

Existing Projects (Brownfield)

Already have code? Start here instead.

1. Map the codebase

/rrr:map-codebase

Spawns parallel agents to analyze your code. Creates .planning/codebase/ with structured analysis of your stack, architecture, conventions, and concerns.

2. Initialize and build

/rrr:new-project

Same flow as greenfield, but the system knows your codebase. Questions focus on what you're adding/changing. Then plan → execute as normal.

The codebase docs load automatically during planning. Claude knows your patterns, conventions, and where to put things.

Why It Works

Context Engineering

Claude Code is incredibly powerful if you give it the context it needs. Most people don't.

RRR handles it for you:

Size limits based on where Claude's quality degrades. Stay under, get consistent excellence.

XML Prompt Formatting

Every plan is structured XML optimized for Claude:

<task type="auto">
  <name>Create login endpoint</name>
  <files>src/app/api/auth/login/route.ts</files>
  <action>
    Use jose for JWT (not jsonwebtoken - CommonJS issues).
    Validate credentials against users table.
    Return httpOnly cookie on success.
  </action>
  <verify>curl -X POST localhost:3000/api/auth/login returns 200 + Set-Cookie</verify>
  <done>Valid credentials return cookie, invalid return 401</done>
</task>

Precise instructions. No guessing. Verification built in.

Subagent Execution

As Claude fills its context window, quality degrades. You've seen it: "Due to context limits, I'll be more concise now." That "concision" is code for cutting corners.

RRR prevents this. Each plan is maximum 3 tasks. Each plan runs in a fresh subagent — 200k tokens purely for implementation, zero accumulated garbage.

No degradation. Walk away, come back to completed work.

Atomic Git Commits

Each task gets its own commit immediately after completion:

abc123f docs(08-02): complete user registration plan
def456g feat(08-02): add email confirmation flow
hij789k feat(08-02): implement password hashing
lmn012o feat(08-02): create registration endpoint

[!NOTE] Benefits: Git bisect finds exact failing task. Each task independently revertable. Clear history for Claude in future sessions. Better observability in AI-automated workflow.

Every commit is surgical, traceable, and meaningful.

Modular by Design

• Add phases to current milestone
• Insert urgent work between phases
• Complete milestones and start fresh
• Adjust plans without rebuilding everything

You're never locked in. The system adapts.

Commands

Core Workflow

Navigation

Verification

Brownfield

Phase Management

Milestones

Session

Utilities

Troubleshooting

Commands not found after install/update?

• Restart Claude Code: type exit, then run claude again
• Verify files exist in ~/.claude/commands/rrr/ (global) or ./.claude/commands/rrr/ (local)
• Still seeing "Unknown skill"? Reinstall with npx projecta-rrr@latest, then restart claude

Commands not working as expected?

• Run /rrr:help to verify installation
• Re-run npx projecta-rrr to reinstall

Updating to the latest version?

npx projecta-rrr@latest

Using Docker or containerized environments?

If file reads fail with tilde paths (~/.claude/...), set CLAUDE_CONFIG_DIR before installing:

CLAUDE_CONFIG_DIR=/home/youruser/.claude npx projecta-rrr --global

This ensures absolute paths are used instead of ~ which may not expand correctly in containers.

Star History

License

MIT License. See LICENSE for details.