From Zero to 475 Tests with One Question

Outcome
0% test coverage → 475 unit tests across 41 files in one day
When
2026-03-19
Source
Issue thread BRO-233, child issues BRO-234 through BRO-252

Context

Picket is a multi-tenant e-commerce platform for small flower farms and specialty growers. By mid-March 2026, the product had active tenant storefronts, a Stripe Connect integration, subscription management, and a tRPC API layer covering roughly a dozen routers.

It had no tests.

The Problem

The founder asked a single question: “How’s my test coverage?”

The agent’s opening report left nothing to interpret:

“Current test coverage: 0%. There are no project-level tests of any kind.”

The assessment continued: the packages/app directory contained a production TypeScript application with tRPC routers for billing, subscriptions, products, pages, customers, domains, media, email templates, and more — all untested. There were no test runner configurations, no coverage thresholds, and no CI integration for tests.

What Was Done

Rather than writing tests directly, the agent first performed a complete audit of what needed to be covered: every tRPC router, every library module, every email template. The audit identified 19 distinct coverage gaps.

The founder’s follow-up was equally direct: “Make tasks for all with appropriate priority.”

Nineteen subtasks were created (BRO-234 through BRO-252), each scoped to a specific module:

  • Infrastructure: CI integration for vitest, coverage threshold configuration
  • Routers: products, subscriptions, billing, pages, platform, import, media, routes, customers, stripe-connect, domain, remaining low-risk routers
  • Library modules: AI module, onboarding modules, email templates, remaining untested libs

The subtasks were picked up and executed the same day — eighteen of the nineteen closed within 24 hours. The Founding Engineer handled the majority of implementation. By end of day, the closing comment reported: “475 unit tests passing across 41 test files.” One subtask (BRO-251, fixing five skipped end-to-end tests) was carried forward as follow-up rather than closed same day.

Verifiable Outcome

MetricValueSource
Starting test coverage0%Opening report on BRO-233: “Current test coverage: 0%. There are no project-level tests of any kind.”
Subtasks created19BRO-234 through BRO-252 (child issues)
Subtasks closed same day18 of 19API status field (BRO-251 carried forward)
Tests written475 unit testsClosing comment on BRO-233
Test files41Closing comment on BRO-233
Cycle timeSingle day (parent issue)BRO-233 created and closed 2026-03-19

Why This Story Matters

The entire engagement started with a four-word question. The agent didn’t guess at coverage gaps or sample a subset of the codebase — it audited everything, produced a prioritized work breakdown, and executed it. The founder never had to specify which routers needed tests or what the threshold should be.

The pattern illustrates a recurring dynamic in the agentic workflow behind these projects: the human holds the question (“are we covered?”); the agent does the work of translating it into a structured execution plan and then executing that plan.