What the source did
- Asks Claude to read a part of the codebase deeply and write findings to research.md.
- Has it write plan.md with code snippets, file paths and trade-offs.
- Adds inline notes to the plan and sends Claude back, repeating 1–6 times with an explicit 'don't implement yet'.
- Asks for a todo list, then one standing 'implement it all' prompt that runs typecheck continuously.
“never let Claude write code until you’ve reviewed and approved a written plan”— Boris Tane, Blog
Summary written by Agent Cookbook. See the original for the full context.
Their prompt
From Boris Tane Their exact wording, one prompt per phase. Source ↗
1. Research
read this folder in depth, understand how it works deeply, what it does and all its specificities. when that's done, write a detailed report of your learnings and findings in research.md
2. Plan (his example)
the list endpoint should support cursor-based pagination instead of offset. write a detailed plan.md for how to achieve this. read source files before suggesting changes, base the plan on the actual codebase
3. Annotate (after adding your notes to plan.md; repeat as needed)
I added a few notes to the document, address all the notes and update the document accordingly. don't implement yet
4. Todo list
add a detailed todo list to the plan, with all the phases and individual tasks necessary to complete the plan - don't implement yet
5. Implement
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.
Approval rule
🛡
The agent must not touch code until you've signed off on plan.md. If it goes wrong, revert with git and narrow the scope rather than patching.
Set it up
- Start a fresh session on a clean branch.
- Run phase 1 and read research.md yourself; correct anything wrong.
- Run phase 2, then annotate plan.md in your editor.
- Repeat phase 3 until you'd approve it as a PR description.
- Run phases 4–5, review the diff, and commit.
Common failures
⚠
- Skipping the 'don't implement yet' guard: the agent starts coding after round one.
- Thin research leads to plans that ignore existing layers. Ask for depth explicitly.
- Long sessions drift; point it back to plan.md, which survives compaction.
Variations
- Paste a reference implementation from another repo with the plan request.
- Use a second agent to review plan.md before you annotate.
Get the 10 best agent setups. Free.
Plus new recipes every week. One email, no spam.
Unsubscribe anytime. Privacy