shipwright

Shipwright

An opinionated, end-to-end development method for Claude Code — and the orchestrator that runs it.

v0.1 — ready to try. The Method + the /ship orchestrator (skill + five lifecycle hooks) are built, tested (72-assertion suite), and live-fired. Install below and take it for a spin — feedback via issues is very welcome. See docs/ for the specs and decision log.

Modern Claude Code has a pile of excellent but disconnected capabilities — superpowers (brainstorming, TDD, worktrees, plan-writing, verification), plan gates (overengineering / second-opinion review), a PR + bot-review cycle, subagent reviewers, an architecture-review CI gate. Used well, they already add up to a near-complete way of shipping software. Used ad hoc, the discipline leaks.

Shipwright is the connective tissue. It names one disciplined loop and gives you something that conducts it:

Brainstorm → Plan → Plan-gate → Implement (TDD) → Self-verify
  → Internal review → Open PR → Automated review → Address → Merge

Plus a session-level Orient (context load) on start/resume, and optional Deploy + Track add-ons after merge.

…with change tiers (Trivial / Standard / Major) so a typo fix doesn't pay the cost of a new-subsystem design, and graceful degradation so the loop still runs when you don't have the full golden-path toolchain.

What this is (and isn't)

• It conducts, it doesn't replace. Shipwright sits on top of superpowers and the Smirnov Labs claude-skills marketplace. It invokes those skills at the right phase — it does not reimplement TDD, worktrees, or brainstorming.
• Golden path + graceful degradation. The full stack (GitHub + a second-opinion reviewer + an architecture-review CI gate + an issue tracker) is the golden path. Missing a piece degrades that phase rather than breaking the loop.
• Opinionated, tiered rigor. The discipline is the product — but tiered, so it's sustainable on real work.

Install

# 1. The frameworks /ship conducts (one-time)
/plugin marketplace add obra/superpowers-marketplace
/plugin install superpowers@superpowers-marketplace
/plugin marketplace add smirnov-labs/claude-skills
/plugin install plan-utilities@smirnovlabs-claude-skills
/plugin install pr-utilities@smirnovlabs-claude-skills

# 2. Shipwright itself
/plugin marketplace add smirnov-labs/shipwright
/plugin install shipwright@shipwright

Then in any repo: /ship — it classifies the change, conducts the loop, and its hooks keep the gates honest. Optional extras: skills-directory/skill-codex (Codex second opinions), andylizf/nonstop (autonomous mode).

The 6 modules

#	Module	Status
1	The Method — the explicit loop + principles + tiers (the contract everything else implements)	✅ spec v1 → docs/specs/2026-06-07-the-method-design.md
2	The Orchestrator — the /ship skill + lifecycle-hook suite that conducts and enforces the loop	✅ built (v0.1) → docs/specs/2026-06-10-orchestrator-design.md
3	Reviewer roster — portable agents + a generator that scaffolds per-surface reviewers from a repo's own conventions	⏳ planned
4	Architect-Gate (generalized) — the architecture-review CI gate, de-coupled from any one project	⏳ planned
5	Adoption flow — one-command init that wires a new or existing repo	⏳ planned
6	Docs + brand — the methodology writeup and identity	⏳ planned

Dependencies it orchestrates

• superpowers (3rd-party) — the core development-discipline skills.
• nonstop, skill-codex — autonomous mode + Codex second opinions.
• Smirnov-Labs/claude-skills — git-recon, detect-overengineering, codex-plan-review, reading-bot-pr-reviews, /pr, /pr-fix, Architect-Gate, Memory Bank.

Status & roadmap

Decisions and rationale are logged in docs/DECISIONS.md. v0.1 is live — current focus: real-world feedback on /ship, then the reviewer-roster generator (Module 3).

---

Built by Smirnov Labs. Not affiliated with or endorsed by Anthropic. "Claude" and "Claude Code" are trademarks of Anthropic.

License

MIT — see LICENSE.