CafeKit runtimefor verified AI coding
Turn ad-hoc prompts into a delivery flow with specs, task boundaries, verification gates, and state sync. Built for Claude Code, with OpenCode support.
npx @haposoft/cafekitFrom first prompt to verified release
CafeKit installs an operating layer into your repo: create specs, split task packets, implement one task at a time, run test/review, sync state, and hand off release through your existing stack.
Install the runtime into your repo
Bootstrap CafeKit into a Claude Code or OpenCode project without touching application code.
Installs skills, agents, hooks, rules, statusline, and workflow context into `.claude` or `.opencode`.
Writes version metadata and an ownership manifest so future installs preserve user edits.
Moves the repo to a spec-driven command surface instead of relying on long prompts.
Run `npx @haposoft/cafekit` in the project root.
The repo is ready to run `hapo:*` workflows with a project-local runtime.
Differentiators
CafeKit does not try to make agents smarter with longer prompts. It puts them inside a runtime with state, artifacts, and gates so progress cannot fake green.
Without CafeKit
Every session needs the same context, scope, and approach explained again
Specs can drift away from implementation after the first coding pass
Tasks can be marked done while build, test, or runtime proof is still missing
Docs are often pushed to the end, partially updated, or skipped
With CafeKit
Specs become artifacts with validation, task registry, and readiness gates
Implementation moves through task packets with completion criteria and Evidence
Quality gates block fake progress before state is synced to done
Docs checkpoints run after verified tasks instead of piling up at release time
What CafeKit installs into your repo
The installer only adds workflow/runtime infrastructure under `.claude` or `.opencode`. Application code stays untouched until you run a workflow for a real feature.
.claude/skills
Skills that create the command surface for question, brainstorm, specs, develop, test, review, docs, sync, and git.
.claude/hooks
Hooks for state/spec drift reminders, sensitive-read blocking, session context injection, and usage updates.
.claude/agents
Subagents for inspect, spec, develop, test, review, docs, git, and deployment handoff.
.claude/rules
Operating rules that keep scope, spec state, quality gates, docs sync, and Git handoff aligned.
Your app code
Left unchanged during install. CafeKit touches app code only when you assign a clear implementation task.
Install the runtime, run the first flow
Install CafeKit, create a validated spec, implement a task with Evidence, then test/review before Git handoff.
# 1. Install CafeKit$ npx @haposoft/cafekit# 2. Create and validate a 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. Test and review the candidate$ /hapo:test --full$ /hapo:code-review --pending# 5. Commit, push, and hand off release$ /hapo:git commit$ /hapo:git push$ vercel --prod
New to CafeKit?
Go from an untouched repo to your first feature with spec, task, and verification receipt.
Real Artifacts after every step
CafeKit leaves files, registries, and receipts so reviewers can see which tasks are ready and which still lack evidence.
Open documentationMachine state for phase, validation status, task registry, and readiness gate.
Task packets with objective, scope boundary, completion criteria, and Evidence commands.
Structured verdicts that show whether code is ready for merge, Git handoff, or release.
Get in touch
Tell us about your project or ask anything. We usually reply within one business day.