CafeKit cup logo
CafeKit
Spec-driven runtime
Claude Code-first runtime for spec-driven delivery

Claude Code-firstCafeKit runtime

Build from approved specs, execute one verified task at a time, and keep project docs aligned. Claude Code is available now.

Claude Code available now
npx @haposoft/cafekit
Spec gates
validate first
Task state
registry-backed
Docs
checkpoint per task
~ /cafekit-runtime
Install
Create spec
Validate
Develop
Test
$
Active Loop
specs → develop → test
Task Packet
R0-02 auth setup

One runtime from first prompt to release

Install the runtime, move through specs, develop, test, review, git handoff, then ship with your existing deploy stack.

Runtime Workflow

Install the runtime bundle

Bootstrap CafeKit into an existing Claude Code project before any feature-specific work begins.

Active Loop
install → specs → develop → test → review → git → deploy
What Happens

Installs skills, hooks, rules, statusline, and workflow context into .claude.

Sets the project up for spec-driven commands instead of ad-hoc prompting.

Keeps Claude Code as the primary supported runtime today.

Critical Notes
Command

Run `npx @haposoft/cafekit` in the project root.

Result

The repo is ready for `hapo:*` workflows immediately.

Why this feels different

CafeKit is most compelling when people can see the difference between ad-hoc prompting and a runtime with state.

Without CafeKit

Long prompts that need to be re-explained every session

Specs drift away from implementation after the first coding pass

Tasks look done in markdown even when builds or tests are still red

Docs updates happen late, inconsistently, or never

With CafeKit

Specs become runtime artifacts with validation and task boundaries

Implementation moves through one verified task packet at a time

Quality gates block fake progress before state is synced

Docs checkpoints happen incrementally instead of as a last-minute scramble

What gets installed

New users want to know whether CafeKit changes the app or just installs workflow infrastructure. Show that clearly.

.claude/skills

The runtime command surface for specs, develop, test, review, git, and sync.

.claude/hooks

State reminders, privacy guardrails, and runtime context injection around Claude Code.

.claude/agents

Specialist agents for testing, review, docs, sync, and workflow enforcement during delivery.

.claude/rules

Execution rules that keep spec state, quality gates, docs sync, and Git handoff aligned.

Your app code

Untouched until you intentionally enter the spec and task workflow for a real feature.

No framework lock-in. Works inside an existing repo and tells you what it is doing.

Get Started in Seconds

Install CafeKit, create and validate a spec, implement a task, verify it, then commit and deploy with your existing release stack.

# 1. Install CafeKit$ npx @haposoft/cafekit # 2. Create and validate a feature spec$ /hapo:specs Build a user authentication system$ /hapo:specs --validate user-authentication # 3. Implement one task packet at a time$ /hapo:develop user-authentication task-R0-02-auth-setup-dual-mode.md # 4. Verify and review the release candidate$ /hapo:test --full$ /hapo:code-review --pending # 5. Commit, push, and deploy$ /hapo:git commit$ /hapo:git push$ vercel --prod

See the artifacts, not just the commands

A new user trusts CafeKit faster when they can picture the files and receipts it leaves behind.

Open documentation
specs/meet-transcript-mvp/
spec.json

Registry-backed state, validation status, and implementation readiness in one place.

task-R*.md

Task packets with objective, completion criteria, and exact verification commands.

review + test verdict

Structured signals that explain whether work is actually ready to merge and ship.