Your First Month with Claude Code

For: Workshop participants who completed the installation and first-project exercises and now want to build a real habit.

The pattern that determines success or failure: In the first four weeks, either Claude Code becomes part of your actual research work — files you care about, questions you actually have — or it becomes something you tried once and will get back to. The difference is almost never skill level. It is whether you used it on something real.

This guide gives you one concrete action per week. Each action is small enough to complete in 30–45 minutes. Together they move you from "I got it working" to "I use this regularly."

Week 1 — One real task on your actual files

Do not use test files or the workshop repository. Use your current project.

Pick one task from the list below and do it this week:

If you have PDFs you need to read:

Convert [paper].pdf to markdown. Summarise the argument in 3 sentences.
What does it say about [your specific topic]? What does it not address?

If you have a draft chapter or section you are working on:

Read [chapter_draft].md.
What is the central argument? Is it clearly stated?
Where is the evidence weakest?
Do not rewrite anything — just diagnose.

If you have notes from a recent meeting, seminar, or archive visit:

Read these notes: [notes_file].md
What are the three most important things to follow up?
What questions do these notes raise that I have not yet addressed?

The specific task does not matter much. What matters: it is your material, your question, your project. If the output is useful, write down what you did so you can repeat it. If it is not useful, note why — that is information too.

End of week 1 check: Did Claude produce anything you actually used or will act on? If yes, the setup is working. If no, try the A.issue.bounding.md diagnostic — the most likely problem is that the task was underbounded.

Week 2 — Write a proper CLAUDE.md for your current project

If you created a minimal start.md during the workshop and converted it to CLAUDE.md, this week is for improving it. If you have not yet created a CLAUDE.md for your current project, create one now.

Start with this prompt:

I want to create a CLAUDE.md for my current research project.
Ask me five questions about the project that would help you work more effectively on it.
Then draft a CLAUDE.md from my answers.

Answer Claude's questions honestly and specifically. Then read the draft and revise it. Typical things to add that Claude's draft will miss:

A CLAUDE.md does not need to be long. One page is right. Two pages is the maximum before parts start being ignored. Every sentence should be something you will want Claude to know every session for the foreseeable future.

End of week 2 check: Launch Claude Code in your project folder. Type: "Read CLAUDE.md and tell me in two sentences what my project is." If the answer is accurate and specific, the CLAUDE.md is working. If the answer is generic, the file needs more specificity.

Week 3 — Do a batch task

This week, do something that would have been painful without Claude Code — a task involving more than one or two files, or something you would have postponed because of the volume of work.

Options:

Reading triage: You have a folder of papers or sources you have been meaning to engage with.

Read all markdown files in /reading_queue/
For each one:
1. One-sentence summary of the main argument
2. Relevance to my project (high / medium / low / not relevant)
3. One question it raises for my work

Write results to reading_triage.md — one section per file.

Note consolidation: You have notes scattered across multiple files.

Read all files in /notes/
Identify: recurring themes across files, open questions that appear in multiple places,
contradictions or tensions I should address.
Write a synthesis to notes_synthesis.md — organised by theme, not by source file.

Reference extraction: You have a folder of papers and need a bibliography or source list.

Read all markdown files in /sources/
For each one: extract the full bibliographic reference (author, title, journal/publisher, year, DOI if visible).
Format as a numbered list in bibliography.md.
Mark [?] where information is missing or uncertain.

Run the task and inspect the output. Look for the failure modes: items that are wrong, items that are missing, items where Claude filled in what it thought you wanted rather than what the source contains. These are normal; catching them is the skill.

End of week 3 check: Did you save significant time compared to doing this manually? Did you catch any errors before acting on the output? If you answered yes to both, the workflow is working. If not, read the failure mode diagnostic in A.issue.prompt-iteration.md.

Week 4 — Evaluate and adjust

This week, do not set a new task. Instead, assess what you have built.

Review your CLAUDE.md. Is it still accurate? Has your current focus shifted in the last three weeks? Are there conventions you established in week 1 sessions that should be added? Remove anything no longer true; add what has become consistently true.

Look at your output files. Read reading_triage.md, notes_synthesis.md, or whatever you produced. How much of it did you actually use? What did you correct? What do the corrections reveal about how to prompt better next time?

Locate yourself on the adoption spectrum. Read B.adoption-spectrum.md or visit the level pages. Based on what you did this month, where are you actually?

Read the leveling package for your current level. C.leveling-packages.md has specific transition guides. The Level 1 → 2 package is the most important for most participants at this point.

The most common reasons adoption fades

Using it on invented tasks rather than real ones. The workshop used example files. If you continued with example files after the workshop, you practiced without learning whether Claude is useful for your actual work. The only test that matters is real use.

Sessions that start from scratch every time. If you are re-explaining your project at the start of every session, you do not have a working CLAUDE.md. Fix this in week 2 — it compounds: every session you delay the fix, the previous session's effort is partially wasted.

Giving up when the first output is not good. Bad first outputs are normal. The skill is diagnosing why and making a targeted fix, not expecting the first prompt to work. Read A.issue.prompt-iteration.md when this happens.

Waiting for a large enough task to justify the overhead. There is no task too small for Claude Code. A five-minute task that Claude does in thirty seconds is still a win; it also builds the habit. Do not save it for "important" work.

Using it only for tasks you do not care about. If Claude only touches your rough drafts and your peripheral reading, it will never feel essential. Give it something you actually need — a chapter section, a source you have been avoiding, a question you are stuck on.

After month 1

If you completed four weeks of real use, you have enough experience to know what Claude Code is actually useful for in your work — not in general, but for you specifically. That knowledge is more valuable than any set of general principles.

The next step depends on what you found:

Related