https://boristane.com/blog/how-i-use-claude-code/

[utterances-bot]

How I Use Claude Code | Boris Tane

The research-plan-implement workflow I use to build software with Claude Code, and why I never let it write code until I've approved a written plan.

https://boristane.com/blog/how-i-use-claude-code/

[abeixiaolu]

Learned a lot, thank you for your selfless sharing.❤

[Rjshakya]

Thanku, boris. As always very helpful ❤️.

[GiffenGood]

I presume you feed the research.md to the planning phase?

[dragosroua]

That’s basically Assess - Decide - Do, with slightly different names, so I can thoroughly relate. I already integrated ADD into my Claude apps, first as a mega prompt, then as a flexible collection of skills and even a Cowork plugin. Feel free to to have a look at the repos, if you think this is helpful.

[GiffenGood]

I really enjoyed this article. It is largely how I work with claude.

Is there any reason you don't write the results of your research to path specific claude.md files, so that:

1. you can re-use in future sessions
2. you are not wasting tokens researching things over and over again

[wgraham17]

I’ve come to similar conclusions and ended up codifying it outside of my IDE/Claude Code (as I’ve had some spare time lately). Proper steering and gated checkpoints are crucial for anything but the most trivial of tasks.

[JoshuaWellbrock]

Nice article, especially the point that using a plan dramatically improves an agent’s output quality. I’ve seen the same effect in practice.

Many of those "manual" steps are also a perfect fit for skills, where you can enforce things like waiting for explicit human approval before continuing.

That’s exactly why we introduced Agile V, an open standard for AI-augmented engineering, along with ready-to-use skills that make planning, approvals, and structured execution easy to adopt.
https://github.com/Agile-V/agile_v_skills

[boristane]

@GiffenGood

I presume you feed the research.md to the planning phase?

I don't always, but if in the plan the agent goes in a direction I don't like, I can directly reference the research.md file in the notes for the agent and steer it in the right direction :)

[boristane]

@GiffenGood

I really enjoyed this article. It is largely how I work with claude.

Is there any reason you don't write the results of your research to path specific claude.md files, so that:

1. you can re-use in future sessions
2. you are not wasting tokens researching things over and over again

Codebases change rapidly, I don't think persisting instructions permanently is the way forward. I have CLAUDE.md files, but they capture only things that are very unlikely to change (frameworks, auth, etc.). I let the agent rediscover everything every time - it's not that costly nor slow.

[alamehan]

Just use GSD (https://github.com/gsd-build/get-shit-done). It automates this exact workflow.

[robinunlearning]

Hey Boris. Great article and insight. Im wondering, what do you do with all the research, plan and todo files? Do you integrate them with a backlog management tool for example?

Cheers
Robin

[mcandea04]

Congrats, you are doing your own flavor of spec-driven development.

[boristane]

Hey Boris. Great article and insight. Im wondering, what do you do with all the research, plan and todo files? Do you integrate them with a backlog management tool for example?

Cheers Robin

@robinmatthews at the moment I delete them once I finish the session with the model. However I intent to start saving them alongside the PR such that they can be referenced in the future, or at code review (although I am very light on code review with this workflow).

[rowanu]

What's so bad about Claude's plan mode @boristane?

[JoanComasFdz]

The superpowers skills are now part of the official Anthropic repository and they do an excellent job, brainstorm for research, write plan for the tasks, execute plan.

I've had a lot of success since I started using them.

On top I use Serena and Context7, it's a great combo.