Build a dev-workflow generator with test-driven agent coding

The author built an open-source system that generates project-specific dev workflows from an interview, entirely with Claude Code, which wrote and debugged 550+ tests.

What They Did

The author says they built the whole system through Claude Code sessions: after a /bootstrap interview about a project's stack, it generates project-specific slash commands, guard hooks, pre-push checks, and agent configs. Claude Code also wrote the main code files, produced and debugged 550+ tests in a test-driven workflow, ran self-audits with four parallel review agents, and, via /task-hunter plus Backlog.md CLI, picked tasks, implemented changes, committed, and closed them autonomously.

How It Went

The result was an open-source, MIT-licensed workflow system the author says spans 15k+ lines, 200+ tasks, 20+ hook files, and supports many common web stacks plus a generic mode. The stated shortcoming was that Claude Code needed guardrails: without hooks, task structure, and automated review, it wrote to wrong directories, lost context, and let subtle bugs through.

Worth Knowing

To try it as described, you need both Claude Code CLI and Backlog.md CLI, then point the template at your codebase and run /bootstrap.