RepCheck Engineering
Building civic tech
with AI agents
Can AI coding agents produce production-grade software if you give them enough context? This blog documents the attempt — the process, the decisions, the failures, and what we learn along the way.
Latest posts
Guardrails That Work: Tagless Final, Integration Tests, and Teaching an Agent to Check Its Own Homework
I refactored bill-identifier to tagless final, wrote integration tests against real PostgreSQL, and discovered that the shell functions forcing local CI checks are the most effective agent guardrail I've built. Here's why constraining the agent made us both more productive.
Designing a 29-Table Database Schema With an AI Agent (And Why I Had to Keep Correcting It)
I asked Claude Code to design RepCheck's database schema. It produced something reasonable, and then I spent an entire session reshaping it through corrections, questions, and design decisions the agent could not make on its own. Here is what the collaboration actually looked like.
Code in the Dark: What Happens When an Agent Can't Validate Its Own Work
I let Claude Code work overnight from my phone while I slept. The code it produced looked right, felt right, and did not compile. Here is why, plus the testability pattern and storage architecture rethink that came out of fixing it.
Almost Ready: Closing the Last Gaps and Templating for Scale
We closed Gap #10, added a giter8 template so every new repo gets the same agent context out of the box, and discovered we were nearly done all along. Here's what the final push looked like.
Closing the Gaps: Shell Battles, Token Economics, and the Last Mile to Agent-Generated Code
We're 8 out of 10 gaps closed on our agent-ready documentation. But getting here meant fighting misconfigured shells, discovering token-saving strategies, and adding the kind of tooling that makes agents trustworthy. Here's what it took.
When Your AI Agent Hits a Wall: Token Limits, API Costs, and the Documentation Tax
What happens when you build a system that depends on LLMs and then discover how much they actually cost? A candid look at the real economics of agent-driven development.
The Real Cost of "Agent-Ready" Code: A Step-by-Step Guide to Building Context for AI Coding Agents
How much work does it actually take before an AI agent can write good code independently? A lot more than you think. This is a practical guide based on a real project — still in progress.
Introducing RepCheck: An Experiment in Agent-Driven Software Development
Can AI coding agents produce production-quality software if you give them enough context? We don't know yet. But we're building the structure to find out.