An Effective Claude Code and Cursor Workflow: Planning, Small Diffs, Review and Tests First

Gushwork expects you to be good with Claude Code or Cursor before day one, and ProcureYard lists AI coding tools as a core skill. Being good with these tools does not mean writing clever prompts. It means steering the agent toward the right change and verifying what it produced, quickly and reliably. Engineers who skip both end up with large, plausible-looking diffs they cannot defend in review. This section gives you a repeatable loop that keeps you in control.

1. The loop

1. Explore. The agent reads the relevant code. You confirm it understood.
2. Plan. The agent proposes a plan. You edit it before any code is written.
3. Test first. Write failing tests that define done.
4. Implement. Take one small step at a time until the tests pass.
5. Review. Read the diff as if a junior colleague wrote it.
6. Commit. Make a checkpoint, then repeat for the next slice.

Each pass through the loop should produce a diff you can review in a few minutes. If it cannot, the slice is too big.

2. Give the agent durable context

Agents start every session knowing nothing about your project. Write the conventions down once.

• Claude Code reads a CLAUDE.md file at the repo root. Running /init generates a first draft for you to edit.
• Cursor reads project rules from the .cursor/rules directory.

Keep the file short and concrete: