What is CAP Pro?

CAP Pro (Code-as-Plan, Pro edition) is a developer framework for AI-assisted coding that follows the Code-First principle: instead of writing requirements documents before touching code, you build a working prototype, annotate it with two lightweight tags, and let the framework derive your project plan from what you actually built.

It works with 8 AI coding runtimes — Claude Code, Gemini CLI, Codex, GitHub Copilot, Cursor, Windsurf, OpenCode, and Antigravity — from a single npx cap-pro@latest install.

The problem CAP Pro solves

Most AI coding frameworks follow a plan-heavy workflow:

discuss → requirements → plan → execute → verify → review
   (9+ steps, with sub-steps and handover docs at every transition)

This produces a stack of planning artefacts — ROADMAP.md, REQUIREMENTS.md, STATE.md, VERIFICATION.md, MILESTONES.md — that drift from reality the moment code changes. You spend more time maintaining documents than building software.

CAP Pro inverts that:

Traditional frameworks	CAP Pro
9+ step workflow	5 steps: brainstorm, prototype, iterate, test, review
8+ mandatory tags / annotations	2 mandatory tags: @cap-feature, @cap-todo
Manually maintained ROADMAP.md	FEATURE-MAP.md auto-derived from code + brainstorm
Separate VERIFICATION.md	Green tests = verified
MILESTONES.md with status tracking	Git tags = milestones
Runtime dependencies	Zero runtime dependencies (Node.js built-ins only)
Single AI runtime	8 runtimes from one install

Tests are verification. Git tags are milestones. Code is the plan.

CAP Pro eliminates accidental complexity by making code the single source of truth.

Aligned with Modern Software Engineering

CAP Pro is aligned with Dave Farley's "Modern Software Engineering" philosophy — optimise for learning and managing complexity. The framework actively encourages:

• Small, frequent integration — every prototype iteration is a learning step.
• Reversibility over commitment — cap-historian lets you fork and roll back any session.
• Empiricism over speculation — RED-GREEN tests, not whiteboard plans, decide what is "done".
• Cognitive load reduction — two tags, five workflow steps, nine focused agents. No more.

The 5-step workflow at a glance

  brainstorm  →  prototype  →  iterate  →  test  →  review
       │             │            │          │         │
  FEATURE-MAP   @cap-tags    scan + fix   green=done  ship

Each step has a slash command (/cap:brainstorm, /cap:prototype, …) and an underlying agent. In normal use you don't need to type the commands — CAP Pro recognises the workflow moment and auto-triggers the right agent. The commands exist as power-user explicit triggers.

See the full workflow guide →

What's in the box

After npx cap-pro@latest, you get:

• 5 per-feature agents — cap-brainstormer, cap-prototyper, cap-validator, cap-debugger, cap-designer
• 4 project-wide agents — cap-historian (snapshots & forks), cap-curator (status & dashboards), cap-architect (system review), cap-migrator (migrations)
• 20+ slash commands — see the command reference
• Hooks — auto-detection, drift checks, status-line integration
• Project memory — .cap/memory/ auto-tracks decisions, pitfalls, patterns, hotspots
• Context7 integration — always-current library docs via npx ctx7@latest
• Multi-user handoff — built-in primitives for forward + reverse handoffs between team members

Where to next?

• Just want to start? → Quick Start
• Want to understand the philosophy? → Code-First Principle
• Coming from code-as-plan@7.x? → Migration guide
• Looking for a specific command? → Command reference