Build a project workflow after a guided interview

They used Claude Code to turn a repo interview into a custom workflow system, while the agent wrote, tested, reviewed, and shipped the code.

Done with Claude Code· Advanced· Apr 2026

What they did
They began with a /bootstrap conversation about the project’s stack and used that to generate tailored slash commands, safety hooks, pre-push checks, and agent settings. Claude Code also produced the core source files, followed a test-first process to create and fix more than 550 tests, and checked its own work with four review agents running in parallel. For task execution, it paired /task-hunter with Backlog.md CLI so it could select work, code the changes, commit them, and mark them done with little supervision.
How it went
The finished system is an open-source project under the MIT license, and the author says it grew to 15k+ lines, 200+ tasks, and 20+ hook files. It supports a range of common web stacks plus a generic mode. The main lesson was that Claude Code performed much better with guardrails; when those were missing, it wandered into the wrong directories, lost track of prior context, and allowed small bugs to slip by.
Tips
Set up the structure, review flow, and directory boundaries before letting the agent work autonomously. To reproduce the approach, use Claude Code CLI and Backlog.md CLI together, aim the template at your repository, and start with /bootstrap.

Try it yourself with Claude Code

I want to build a project-specific workflow for [repository or app] using [language or stack]. Start by interviewing me about the codebase, common tasks, and folder structure. Then create custom slash commands, protection hooks, pre-push checks, and agent configuration files for this project. Next, implement the system in a test-driven loop: write a failing test, make the smallest code change needed, rerun the relevant tests, and fix any failures before moving on. If a step would delete files, submit data, make a purchase, or send something externally, pause and ask for my approval first. Continue until [workflow pieces or features] are complete, then give me a short recap of the tests you added, the files you changed, and any risks you noticed.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Read the original on reddit.com ↗

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.