First changes with Claude

How can you use Claude Code on an existing project without losing control of changes?

  • Lovable
  • Claude Code
  • Codex

1250 € per session, excl. VAT

In short

This half-day session of 3 hours introduces participants to using Claude Code in a small existing project. Participants scope a change, provide contextualised instructions, review the proposed code and run tests. They leave with a documented local change and a reusable verification method.

Why this training?

In an existing project, an overly broad coding request can lead to unnecessary changes or conceal a regression. Identifying the relevant files, constraints and expected behaviours helps keep an intervention understandable. Generated code remains a proposal to be checked. The course combines precise instructions, human review and local tests to distinguish a convincing response from a change whose behaviour has actually been observed. Continuing case study: The fictional organisation Avelune Réemploi uses a small local application to track collected items. Its team wants to filter items by status without changing the stored data. The training project contains fictional data, a few tests and a deliberate fault in the filter processing. Transfer to work: On an authorised project, carry out a limited correction: record the initial state, write a bounded instruction, examine the differences and compare tests before and after. Have the change reviewed before any integration into a shared environment.

Objectives of this training

  • Identify the structure and entry points of the supplied project.
  • Write an instruction specifying expected behaviour, scope and exclusions.
  • Obtain and then examine a limited code change.
  • Run tests and compare their results before and after.
  • Document the changes retained and the remaining limitations.

Who is it for and what are the prerequisites?

Roles and tasks concerned

  • Developers starting to use Claude Code on existing projects.
  • Technical professionals able to read and modify a small program who want to practise assisted coding.

Prior skills

  • Have already used a generative AI assistant for a simple task.
  • Be able to read code, navigate files and run a command in a terminal.
  • Understand the role of a test and a version history.

What you will work on

  1. Read the project before intervening

    Objective: Establish an observable initial state. File structure, local startup, existing tests and boundaries of the shared context. Checking secrets and authorisations.

  2. Formulate a bounded request

    Objective: Turn the filtering requirement into a verifiable instruction. Useful context, input and output examples, exclusions and acceptance criteria. Distinguishing between a requested explanation and an authorised change.

  3. Modify, then review

    Objective: Retain only the necessary changes. Generating and modifying code from instructions; reading differences, unexpected dependencies and reverting to the previous state.

  4. Test and prepare the handover

    Objective: Evaluate the result without confusing generation with validation. Standard cases, boundary cases and regression checks; before/after comparison; traceability of faults and limitations.

What you take away

  • Brief project map and record of initial tests.
  • Contextualised instruction with acceptance criteria.
  • Reviewed local change and file comparison.
  • Test report and handover note.

Method and assessment

  • Short explanations followed by immediate practice on the continuing case study.
  • Individual work followed by work in pairs, with peer review of outputs.
  • Guided group discussion: what works, what needs correcting and what remains a human decision.
  • Preparation: Arrange access to Claude Code authorised by the organisation; check installation and terms of use in the official documentation. Download the fictional training project, check that it starts locally and remove any secrets from files accessible during the exercise.

Observable criteria: The instruction specifies the filter's behaviour, the targeted files and the exclusions. ; Every modified file is reviewed and every retained change is justified. ; Tests cover an existing status, no results and a return to all items. ; The report distinguishes between passed tests, failures and checks not performed. ; No secrets or real sensitive data appear in the context provided..

Frequently asked questions

Are Lovable and Codex also used?

Practice focuses on Claude Code. Lovable and Codex are mentioned as alternatives only; their specific arrangements should be checked in their official documentation.

Can participants bring a work repository?

Only if it is authorised, anonymised and free of secrets. The supplied fictional project remains the shared training resource.

Is knowledge of code necessary?

Yes: participants must be able to review changes and interpret tests without delegating their validation to the assistant.

Will the result be deployed to production?

No. The deliverable is a documented local change; any subsequent integration is subject to the organisation's controls.

Last updated: