57 engineering skills activate invisibly while you code. TDD, security, architecture, quality gates — enforced automatically. Context persists across every session. You see everything it does.
npx shipworthy init
Two agents ran identical 7-turn non-technical founder prompts. 36 automated checks — written before either agent ran — scored the output.
Full methodology, all 10 task definitions, and reproduction scripts on GitHub.
Run npx shipworthy init in any project. Hooks install automatically. Skills load on demand. No config files to edit, no workflows to learn.
Six hooks run in real time. Files scanned for secrets before writes. Bash commands checked for destructive patterns. Dependencies monitored. Everything logged transparently to your terminal.
Say "build a payment API" and TDD, input validation, security, and API design skills activate automatically. You code normally. The engineering happens underneath.
Hooks capture every session event — commits, security warnings, pattern detections. A regression fence of "NEVER do X" rules loads every session. No more repeating the same mistakes.
Session signals auto-process into learnings and fence rules at next session start. Corrections become permanent knowledge. Every session starts smarter than the last — 95% automatic.
Each skill is a Markdown file. No code, no runtime, no dependencies. Just engineering instructions that make AI agents build production-grade software.
Six hooks guard every action. Secrets caught before files are written. Destructive commands blocked before execution. Security scans logged transparently.
Failing test first, then implementation, then verification. TDD happens automatically. You see "it works." Underneath: tests prove it.
5 files: light touch. 50 files: coverage thresholds. 500 files: security scans, performance budgets, circular import detection.
Auto-detects application type — web, API, GraphQL, mobile, CLI, pipeline, IoT, desktop, infrastructure — and applies the right security checks.
Hooks capture session signals automatically. Regression fence loads anti-patterns every session. Auto-retro processes corrections into permanent knowledge. 95% automatic.
Every hook, signal capture, and fence check logged with color-coded output. Skills announce themselves before activating. Toggle off with one environment variable.