Spec-driven development for AI coding agents

Turn “describe what you want” into a workflow you can trust.

SpecDevShip is a 19-skill Workflow Pack that wraps your AI coding agent in a real spec-driven pipeline — triage, tier decisions, root-cause investigation, and approval gates that actually hold. Built for solo developers shipping products they'll still be maintaining in a year.

Claude Code Edition — available now More agents — coming soon

pipeline
IdeaTriage & tier decision
Investigate / AnalyzeBug’s root cause — or research on an existing app, design, or product
SpecLight plan, or full requirements→tasks
Approve & buildGate holds; per-task commits
ShipWrap-up, changelog, semver
See the full workflow ↓
The problem

Vibe-coding has a ceiling.

Your AI coding agent will happily generate a confident, plausible plan for the wrong problem at the same speed it generates the right one. Without a system wrapped around it, that speed works against you.

01

No paper trail

Decisions live in a chat that scrolled past. Three weeks later, nobody — including the AI — can tell you whether something was a deliberate call or a guess that happened to compile.

02

No brake pedal

Left alone, an AI agent doesn't reliably tell "add a button" from "redesign the data model." Small requests quietly become big changes mid-implementation.

03

No root cause

“Probably a race condition” is a shrug with technical vocabulary. Fixes get proposed before the actual cause is ever found.

How it works

One pipeline, from idea to shipped.

Every piece of work enters through intake, gets explicitly scoped, and hands off automatically to the next stage — the way a disciplined teammate would, without you having to remember the process yourself.

workflow.mmd
flowchart TD
    A[Idea] --> B{Tier decision}
    B -->|Light tier| L1[spec-plan]
    B -->|Full tier| F1[spec-requirements]
    A -.->|unclear cause| INV[investigate]
    INV -.-> B
    A -.->|existing material| AN[app / design / art / product analysis]
    AN -.-> F1
    F1 --> F2[spec-tech]
    F2 --> F3[spec-design]
    F3 --> F4[spec-tasks]
    L1 --> APR{Approved?}
    F4 --> APR
    APR -->|Yes| IMP[implement-tasks]
    APR -.->|Revise| L1
    IMP --> WRAP[wrap-up]
    WRAP --> REL[prepare-release]
01 Idea Analysis — triage & tier decision 02 Investigate / Analyze — unknown cause or existing material 03 Spec — light plan, or requirements→tech→design→tasks 04 Implement — checklist-driven, per-task commits 05 Wrap-up & Release — walkthrough, changelog, semver
What's inside

19 skills, each doing one job well.

Not a single mega-prompt — a set of specialists that hand off to each other, the way a real team divides work.

Triage

Two tiers, so small stays small

Most work is light tier: one plan, one focused test script. Only genuine new-architecture work escalates to the full requirements → tech → design → tasks pipeline.

Diagnosis

Investigate before you fix

Bug reports get a real root-cause investigation, cited to file and line, before any fix is proposed — never a guess dressed up as a diagnosis.

Analysis

Reverse-engineer before you extend

App, design, art, and product-identity analysis skills extract what actually exists before a redesign or extension gets specced against a half-remembered impression.

Origination

Product, design & art as real roles

A structured Product Manager interview for new ideas; a Designer and Art Director that pitch real directions, get your pick, then run with full autonomy.

Governance

Approval gates that hold

Every spec carries draft/approved status. Nothing gets silently edited out from under you once you've signed off on it.

Shipping

Wrap-up & release, automatically

Walkthroughs, changelog entries, and semver bumps derived straight from your commit history — no separate release ritual to remember.

One workflow, every agent

Built once, works everywhere your agent runs.

The tier decision, the investigation discipline, and the approval gates aren't tied to any one AI coding agent — they're the workflow itself. Editions ship one at a time; this page updates as each one goes live.

Claude Code Available now
Next edition Coming soon
Next edition Coming soon
Next edition Coming soon

Buy once, and every future agent edition of the Workflow Pack is included as a free update — for as long as SpecDevShip keeps this product actively maintained and for sale. See the Terms for the full "lifetime updates" definition.

Right-sized, every time

Light tier for most work. Full tier when it counts.

Light tier (most work)Full tier (rare)
WhenFeatures, fixes, UI work, refactorsNew app, or real new architecture/risk
Outputimplementation_plan.mdrequirements.md → tech.md → design.md → tasks.md
TestingFocused regression script + manual checksCoverage-targeted automated suite
Start here

Read the workflow, then install it.

The field guide walks through the whole system in plain language. When you're ready to run it, the full skill pack drops straight into your own project.

Free field guide $0

A practical eBook covering the tier-decision framework, the investigate-before-you-fix discipline, and how approval gates work day to day.

  • The full workflow, explained end to end
  • The reasoning behind each design decision
  • No pack installation required to read it

Delivered by email. Unsubscribe anytime.

FAQ

Before you ask.

Does this only work with Claude Code?

Right now, yes — the Workflow Pack currently ships as the Claude Code Edition, a drop-in project template plus personal account skills. Editions for other AI coding agents are planned; see the Agents section above for what's currently supported.

If I buy now, do I get future agent editions too?

Yes. Every future edition of the Workflow Pack — for whichever agents we add support for next — is included as a free update to your existing $49 purchase, for as long as this product remains actively maintained and for sale. It doesn't extend to unrelated future SpecDevShip products; full definition in the Terms.

Is this overkill for a small project?

For a one-off script, yes — skip it. It earns its keep on anything you'll still be maintaining in six months, where an undocumented decision or an unfindable root cause actually costs you time.

What's the difference between the free guide and the paid pack?

The field guide explains how the workflow thinks — the tier decision, the investigation discipline, the approval gates. The pack is the thing itself: all 19 skills, installed and running in your own project.

Do I need to follow every skill, every time?

No. Idea Analysis routes you automatically — most work only ever touches the light-tier plan skill. The full pipeline and the origination skills (product, design, art direction) only come up when the work actually calls for them.

Is it a subscription?

No. The skill pack is a one-time $49 purchase with lifetime updates. The field guide is free.

From the blog

Still deciding? Go a layer deeper.

Three places to start — how the tier decision actually works, what shipping an MVP with it looks like, and the full case for ditching vibe-coding.