Project Factory

A Claude Code skills pipeline that guides you from a raw idea to a shipped product.

What Is This?

Project Factory is a set of Claude Code commands and artifact templates that walk you through a structured 9-phase pipeline. It is designed for people who have no software engineering experience — Claude teaches you as you go using direct instruction methodology.

This is not a software product. It is a workflow repo meant to be used directly inside Claude Code.

How to Use

New Projects

1. Open this repo in Claude Code
2. Run /start-project and describe your idea
3. Claude guides you through each phase automatically — explaining, teaching, and asking for approval at every step

/start-project is the recommended entry point. It walks you through the full pipeline. Individual phase commands are available for re-running specific phases or advanced use.

Existing Codebases

1. Open your project in Claude Code (with this repo's commands available)
2. Run /enhance-project
3. Claude audits your codebase, compares it against requirements, discusses priorities, and creates a gap-driven plan

Pipeline Phases

#	Command	Purpose
1	/understand	Interview to clarify your idea
2	/prd	Write product requirements
3	/presearch	Research technology options, plus a blind second opinion from another AI (Codex)
4	/decide	Lock in technology and product decisions
5	/plan	Break work into implementation phases
6	/implement	Build it using TDD, one slice at a time
7	/review	Review implementation quality
8	/test-qa	Evaluate test coverage and quality
9	/ship	Prepare environment, deployment, and release

Learning Commands

Command	Purpose
/teach	Explain the current phase to a complete beginner
/reflect	Full retrospective — decisions, system design, product thinking, and code/implementation details at interview depth, plus interview prep
/interview	Mock interview with 3 questions based on what was built, with feedback

Enhancement Commands

Command	Purpose
/audit	Analyze an existing codebase (technology, features, quality)
/gaps	Compare PRD requirements against audited codebase
/polish-ui	Review frontend for AI-generated patterns and create an authenticity improvement plan

Specialized Commands

These are standalone add-ons that live outside the gated 9-phase pipeline. They read factory/artifacts/ for grounding when it exists.

Command	Purpose
/arch-docs	Generate architecture documentation, a decision log, and a privacy analysis

The four eval commands form a 4-stage AI-evals progression, each building on the previous one:

Command	Purpose
/golden-evals-architect	Stage 1 — a golden eval dataset (ground-truth test cases)
/labeled-scenarios	Stage 2 — a labeled scenario coverage map
/replay-harness	Stage 3 — a deterministic replay harness
/rubric-evals	Stage 4 — LLM-as-judge rubric scoring

Recommended Flow

New Projects

/start-project → (Understand, PRD + appendix, Presearch + blind Codex second opinion, Decide, Plan)
  → build kickoff approval → autonomous background build (worktree agents, TDD, stop-condition pauses)
  → Security report gate → Debug report gate → Reflect (stack + architecture)
  → /review → /test-qa → /ship

/reflect and /interview are available any time for the full retrospective and practice.

Existing Codebases

Same pipeline from the build onward, front-ended for existing code:

/enhance-project → (Requirements, Audit, Gaps, UI Polish, Discussion, Presearch-of-the-delta, Decide, Plan)
  → build kickoff approval → autonomous background build (worktree agents, TDD, stop-condition pauses)
  → Security report gate → Debug report gate → Reflect (stack + architecture)
  → /review → /test-qa → /ship

Artifacts

All pipeline outputs are stored in factory/artifacts/. These files are the source of truth for the workflow — each phase reads from previous artifacts and writes its own.

Artifact	Written By
PROBLEM_SUMMARY.md	/understand
PRD.md	/prd (product-only)
PRESEARCH_QUESTIONS.md	/prd
PRESEARCH.md	/presearch
PRESEARCH_CODEX.md	/presearch (Codex second opinion)
LOCKED_DECISIONS.md	/decide
PROJECT_PLAN.md	/plan
PLAN_FOR_REVIEW.md	/plan (cross-AI review export)
IMPLEMENTATION_LOG.md	/implement
REVIEW_REPORT.md	/review
TEST_REPORT.md	/test-qa
ENV_SETUP.md	/ship
DEPLOYMENT.md	/ship
RELEASE.md	/ship
CODEBASE_AUDIT.md	/audit, /enhance-project
GAP_ANALYSIS.md	/gaps, /enhance-project
UI_POLISH_REPORT.md	/polish-ui, /enhance-project
SECURITY_REPORT.md	security agent
DEBUG_REPORT.md	debug agent
REFLECTION.md	/reflect
LEARNING_NOTES.md	/teach, /reflect

Structure

.claude/
  CLAUDE.md              # Global rules for Claude
  commands/              # Pipeline command files (22 commands)
factory/
  artifacts/             # Pipeline state and outputs (21 artifacts)