My Claude Code Setup (www.justindfuller.com)

A developer published a detailed workflow showing how they use Claude Code to produce 100% of their production, user-facing code without sacrificing quality—while explicitly retaining full responsibility for every line. The post outlines principles (quality over speed, human responsibility), results (comparable-or-better code quality, equal-or-higher JIRA throughput, more time for tech debt), and constraints (terminal-first, not IDE; primary languages: Go and TypeScript). They run 1–4 Claude instances (model pinned to Sonnet 4.5), rely on subagents for specialized reviews, and emphasize careful plan review and iterative checks before letting the model execute changes. Technically, the workflow centers on git worktrees to isolate concurrent Claude sessions, a strict plan-mode/design-review loop, and guarded “agentic” execution with Auto-Accept Edits plus permissions to run tests, linters, builds, and read project files. Custom slash commands (/worktree, /plan, /ship, /review, /github-comments, etc.) automate branch creation, draft PRs, commit messaging, JIRA updates, and GenAI labeling. MCP integrations give controlled browser, JIRA, GitHub, and Drone access for build-log analysis. The significance for AI/ML engineering: it’s a practical, reproducible blueprint for production-grade AI-assisted development that balances automation and human oversight, shows how to embed safety/escape hatches, and highlights real productivity gains while underscoring ongoing human responsibility and limits.