Ship right, not just fast.
Your AI writes the code. You stay in control.
Shipwright is the harness your AI follows, the discipline layer that keeps it true to what you already decided. The full development lifecycle, orchestrated as a sequence of inspectable phases, not a black box. Run it from the CLI, manage it from the Command Center, iterate daily.
The problem
0%
of all code is now AI-generated. Almost none of it is controlled.
“getting lost in the codebase they themselves created.”
HACKER NEWS, 2026
“after two years of vibe coding, I’m back to writing by hand.”
865 UPVOTES, HN
“AI agents do a terrible job without that discipline.”
HACKER NEWS, 2026
The answer isn’t to stop using AI. It’s to put a process around it.
Whitepaper
Driving the AI SDLC with Shipwright.
AI made writing code fast, and nothing checks the result against what you decided. This paper is the thinking behind Shipwright: four layers that close that gap, a baseline that is checked, a harness that enforces, a loop every change goes through, and evidence that can be read. It shows how each one works in practice, and ends with ten questions to ask of any AI-assisted way of building.
PDF · October 2026
The harness
Keeps your project true to what you already decided.
Giving an AI your specs and architecture is solved, most good tools do it now. Keeping that record true as the product changes is the hard part. On every change, Shipwright checks the work back against your baseline and won’t let through a change that silently drops a requirement or reverses a call you already made.
▦
SPECS & GOALS
Requirements
◈
SYSTEM DESIGN
Architecture
⑂
ADRS & PLANS
Decisions
›_
IMPLEMENTATION
Code
How it works
Start /shipwright-run and let the pipeline guide you.
Phase by phase, transparent, learnable. You see exactly what each phase produces before the next begins.
01
Specify
Describe what to build, and Shipwright interviews you into structured IREB specs.
02
Design
Clickable HTML mockups from your specs, brand tokens auto-extracted. No Figma.
03
Plan
Reviewable sections, each tracing back to a requirement.
04
Build
TDD with adversarial code review, checked against the mockup.
05
Test
Up to 8 layers: unit, integration, pgTAP and RLS, smoke, E2E, fidelity, perf.
06
Secure
Semgrep, Trivy, Gitleaks. Auto-fixable findings get patched and retested.
The daily workflow
Not every change needs ten phases. Iterate scales to fit.
Shipwright sizes each change in two stages, a quick estimate from your prompt, then a repo scout that scans the actual files, and runs only the phases that change needs. Then complexity locks.
And every change must trace to a requirement, or it is rejected at write time. Fail-closed, not advisory. That is the part vibe coding skips.
SHIPWRIGHT-ITERATE : Session Plan
Run ID:
iterate-2026-04-05-course-search
Intent:
FEATURE
Complexity:
Small (1 FR, ~4 files, risk: touches_migrations)
Phases:
spec → design → build → review → test → finalize
Skipping:
iterate spec, mini-plan, full review
Safety floor:
DB migration → mandatory down.sql
TRIVIAL
1 FR · 1-2 files
Build, self-review, unit test. Minutes.
SMALL
1-2 FR · 3-5 files
Plus confirmation, mini-plan, conditional review.
MEDIUM
2-4 FR · 5-10 files
Plus scoping interview, external review, full suite, E2E.
LARGE
4+ FR · high risk
Escape hatch, hands off to the full pipeline.
One change. Every artifact updated in lockstep.
Each iterate ends the same way. Spec, decisions, tests, changelog and the compliance reports all move together, or the commit is rejected. Nothing silently drifts.
1 change · committed
Spec / FRs
✓ in sync
Architecture
✓ in sync
ADRs · Decision log
✓ in sync
Tests + ledger
✓ in sync
CHANGELOG
✓ in sync
Compliance · RTM · SBOM
✓ in sync
Event log
✓ in sync
touches_auth
touches_rls
touches_middleware
touches_migrations
touches_billing
touches_shared_infra
cross_split
touches_public_api
touches_build
touches_io_boundary
cross_component
11 risk flags enforce safety minimums, full review, full test suite, or rollback scripts, regardless of complexity.
Command Center
Your AI dev team needs a control room.
A local web interface where Claude moves the cards on a Kanban board, not you. No 20 windows open.
Every task has its lane on the board.
Pipeline and campaign lanes, one card per task. Each card shows where the task stands and launches or resumes it from there.
1
Lanes. Backlog, in progress, done. One card per task.
2
Card. Project, state and run type at a glance.
3
Launch or Resume. One click starts the run or picks it up where it stopped.
The Ship’s Log is the project’s home.
Every run is an entry. The Control Grade sits on top, with the specs, agent docs and compliance reports one click away, and a box to start the next change.
1
Control Grade. The grade on top, with the reason behind it.
2
Next change. Type what you want to change and hit Plan it.
3
Docs. Specs, agent docs and compliance reports, one click away.
The live session streams right on the task.
Launch a pipeline or iterate from any task and the Claude command runs in an embedded terminal on the task page, output streaming in place. The Command Center follows the transcript, it never spawns Claude itself.
1
Files. The project tree next to the session.
2
Live terminal. The Claude command streams right here as it runs.
3
Iterate spec. The plan for this change, open beside the run.
Kanban Task Board
Backlog to Done, one button to start a run.
Global Inbox
Every question Claude asks, in one place.
Multi-project
Many projects, one board.
Free grader
What’s your repo’s Control Grade?
A plain A to F grade for how much you actually control your AI-built code. Traced to requirements? Tested? Reviewed? Reconciled when it changed? Run /shipwright-grade on any repo. A local path never leaves your machine.
Grade your repo
EXAMPLE GRADE
D−
Out of control.
AI changes you can’t verify, a typical vibe-coded repo.
Requirements traced
✗
0%
Tests
✗
11%
Human-reviewed
✗
none
Reconciled on change
✗
drifted
Security scan
✓
passed
See it run
Watch it work, start to finish.
Three walkthroughs, recorded from the real tool.
Command Center walkthrough
Kanban board, embedded terminal, Mission and the Ship’s Log in one connected flow.
Watch on YouTube · 8:50
Intent Wizard: idea to first run
A rough idea and four answers become a Flight Plan you review before work starts.
Watch on YouTube · 4:34
Control Grade: what can your repo prove?
How the A to F grade is calculated, and how to read the report.
Watch on YouTube · 9:36
Masterclass coming soon
Founding seats still available
$49
Founding · first 50
$97
Early access
$497
Full price
Get the founding offer · $49
Lowest price it will ever be. Fully refundable until launch.
Who’s behind it
Ship right, not just fast.
I’ve spent 20+ years on the process of shipping software cleanly, including FINMA-regulated banking and digital-asset environments, where every change comes with proof that it is safe. Shipwright encodes that discipline so you don’t have to carry it in your head.
Honest limits
What Shipwright doesn’t do.
✕
Behavior correctness
Whether it does what users actually need stays your call.
✕
Mechanical, not vibes
Hooks block dangerous actions in code, not in advisory prose.
✕
Not a black box
You stay in the loop at every phase. It does not replace your judgment.
Take control of what your AI ships.
Free, open source. The founding offer for the Masterclass is $49, first 50, fully refundable.
© 2026 Sven Roth