Separating Planning and Execution
https://boristane.com/blog/how-i-use-claude-code/
The author shares a structured methodology that divides the process into stages so that Claude doesn't write code "blindly" but instead works according to an approved plan. Research and planning are always conducted first, and only then comes the implementation. This prevents mistakes, maintains control over the architecture, and minimizes token usage.

Workflow Stages

1. Research Phase: We use words like "deeply" and "in detail" in prompts to review what already exists — the agent documents this in a research.md file.
2. Planning Phase: We create a detailed plan in a plan.md file describing the approach, code snippets, file paths, and trade-offs.
3. Annotation Cycle: We open the editor and add notes directly into the plan (e.g., "use PATCH, not PUT"), then tell the agent: "I added a few notes to the document, address all the notes and update the document accordingly. don’t implement yet". This is done iteratively several times.
4. Todo List: When everything looks good, the agent converts the plan into a detailed checklist of tasks. We continuously remove unnecessary items from the plan to avoid scope creep.
5. Implementation Phase: After the plan is approved, a standard prompt is used: "implement it all," with instructions to mark completed tasks, check types, and avoid unnecessary comments: "implement it all. when you’re done with a task or phase, mark it as completed in the plan document. do not stop until all tasks and phases are completed. do not add unnecessary comments or jsdocs, do not use any or unknown types. continuously run typecheck to make sure you’re not introducing new issues."

Practical Tips: Provide the agent with links to open-source projects that contain examples of similar code. Refer back to the plan when something goes wrong.

Discussion
https://news.ycombinator.com/item?id=47106686
Many users agree with the principle of separating planning and execution, considering it an effective way to reduce errors. Using detailed plans in .md files provides a clear track record of decisions and reasoning. Plans help identify the model's biases, making the process more transparent.

Critics call this method of programming "garbage" or a "gamble," arguing that it leads to both "brain atrophy" due to AI dependency and poor code quality.

#claudecode #prompts