Where we are
You've picked a task worth automating and defined what "done" looks like. Now begins the planning phase. It starts with building your context library.
Planning is where you should spend most of your time
The amount of time you spend on planning directly affects everything downstream: the quality of the execution, the consistency of the output, the ability to create a standard, repeatable process. Some tasks need heavy planning. Some are simpler. But it is always worth thinking about how you're going to tackle the problem before asking Claude to generate output.
Organised context, consistent output, repeatable process you can use again.
Vague context, inconsistent results, starting from scratch every time.
When planning, I generally think about the structure and architecture for the workflow. Then I use Claude Code to give me additional guidance, more ideas, and plug in best practices. It's a mixed planning method: part manual thinking on paper, part AI-assisted exploration.
Think about what Claude needs
Before writing a single prompt, ask yourself: what would the AI agent need to be able to generate the output I want? For our AT4D communication pack, that means a brand-consistent, visually appealing PDF document with all the right information. Taking this into consideration, think about the different inputs and context components you want to give Claude Code.
Think of it as briefing a new colleague. You wouldn't just hand someone a task and hope for the best. You'd give them background reading, brand guidelines, examples of good work, and clear instructions. That's exactly what your context library does for Claude.
The way to organise this is simple: think in folders. Each folder represents a category of context that Claude needs to do its job well.
The context library
There's no single right way to structure your context library. This is how I organise it for this particular task. You can adapt it to fit your style and your project. It needs to work for you.
This is a fairly straightforward context library. You can structure yours differently. The key is that it covers the categories of information your AI agent needs to produce quality output.
If you're not sure how to build your context library or you don't have any context to fill in, you can ask Claude Code in the planning phase to help you structure and populate it. Claude can suggest what folders you need, what kind of content belongs in each, and even help you draft some of the missing materials.
Populating the library with real content
Once the structure is in place, populate it with actual files. The more context you give Claude Code, the better the outcome.
In some cases, you won't have everything you need to populate your library right away. That's fine. Start with the structure. You can always add more later, and Claude can help you fill the gaps.
Create the folder system, even if some folders are empty. The architecture matters more than completeness at this stage.
Drop in what you have. Each file you add improves the quality of Claude's output. More context, better results.
What's next
Your context library is structured and populated. Claude now has the raw materials it needs. In Part 3, you'll move to the next phase of planning: writing the instructions that tell Claude how to use this context. You'll define a persona, map out the tools, design the process sequence, and write the context.md file that ties everything together.