Software Factory
Claude Code and Codex plugin with a team of ten AI agents that take a product from idea to release behind three human approval gates
Overview
Software Factory is a plugin for Claude Code and Codex that gives you a team of ten AI agents: a Team Lead, a Triage Classifier, a Product Manager, a Software Architect, a Developer, a Test Automation Engineer, plus DevOps, FinOps, Security and UX/UI reviewers. The Team Lead triages each request, so a small fix takes a fast path while a new product runs the full lifecycle: discovery and PRD, architecture with design reviews, a per-task build loop, and release readiness. It stops for your approval at three gates (PRD, design and cost, release) and before anything irreversible, such as a push, merge, deploy, infrastructure apply or paid resource. It also works on existing repos, where it maps the codebase and reviews or extends it.
The Verdict
Who Should Use Software Factory?
Best For
- Solo builders who want a PM, architect, tester and security reviewer around their coding agent
- Teams that want human sign-off before pushes, merges, deploys and paid cloud resources
- Claude Code or Codex users who track work as GitHub issues on a project board
Not Ideal For
- Quick one-off scripts where reviews and gates add overhead
- Users of other coding agents (only Claude Code and Codex are supported)
- Organizations that need a licensed, widely adopted project with a support channel
What's Great
- Triage first, so small requests skip the full lifecycle
- "Done" needs command output re-run by the Test Engineer and Architect, not just the Developer's claim
- Bounded build loops: at most three fix rounds per task and three attempts per failing check
- Security (STRIDE, ASVS), FinOps cost review and WCAG 2.2 AA UX review built into design and build
- Routes work to Haiku, Sonnet or Opus by task type to control cost
Watch Out For
- Very new, with little public adoption so far
- No license file in the repository
- Subagents cannot ask you questions directly; everything routes through the Team Lead
- Plugins load only at session start, so open chats will not see a fresh install
- GitHub Project tracking needs the
ghCLI with the project scope
Pricing
View all features & details
Key Features
- Ten agents with a shared handoff contract
- Lifecycle: discovery, PRD gate, architecture, design reviews, design and cost gate, build loop, release gate
- Separate infrastructure gate: DevOps plans, Architect and FinOps review, apply only after approval
- Per-repo "factory brain" of verified facts that agents read
- GitHub issues and project board tracking, with a tasks.md fallback
- Model policy setting: default, conserve or quality
- Project-level agent overrides and per-project agent memory
Use Cases
- Taking a product idea to a reviewed, tested release
- Mapping and reviewing an existing codebase
- Adding a feature with architecture, security and cost review
- Running infrastructure changes behind human approval
How It Compares
| Feature | Software Factory | Mifal | BMAD-METHOD |
|---|---|---|---|
| Approach | Agent team with approval gates | Shared workflow with brief/receipt contracts | Agile persona agents and workflows |
| Agents supported | Claude Code, Codex | Claude Code, Codex, Droid, Pi, OpenCode | Many AI IDEs and agents |
| Human gates | PRD, design and cost, release, plus irreversible actions | Lead plus reviewer must accept receipts | Checkpoints between phases |
| Security, cost and UX reviewers | Built in | Not built in | Not built in |
| License | Not stated | MIT | MIT |
Sources: Software Factory · Mifal · BMAD-METHOD