Skip to content

Claude Code · team setup

A harness for your team's AI coding agents.

Whether your developers are just starting with AI or already use it on their own, I build a shared setup in your repo — instructions, skills, hooks and guardrails — so coding agents work the same way for everyone, and safely.

The problem

AI coding is already in your team. The setup is not.

  • just starting

    Your developers are just starting with AI.

    Everyone tries a different tool and a different prompt. Nobody has decided what the agent may change, what it may read, or who checks its work.

  • already using it

    They already use it, each on their own.

    It works well for some and badly for others. The good instructions live in one person's config, secrets sit one prompt away, and review cannot keep up with what the agents produce.

What a harness is

The shared setup around the agent

A harness decides what the agent knows, what it may do, and what stops it. It lives in your repository, next to the code, so it is the same for every developer and every session — and it changes through review like the code does.

  • CLAUDE.md

    Shared instructions

    How your team builds, tests and reviews, written once and read by every agent at the start of every task.

  • .claude/skills/

    Skills

    Your recurring work — a release, a migration, a review — as tested steps the agent follows, not prompts people retype.

  • hooks/

    Hooks and guardrails

    Checks that run before every tool call. Deleting data, rewriting history or pushing a secret is blocked before it happens.

  • sessions

    Isolated sessions and review

    Each task in its own clean checkout and branch, and nothing merges without a review round.

Services & pricing

Three steps, each at a fixed price

Start with an assessment. The report it ends with is the fixed quote for the build.

Step 1

Assessment

For: Teams deciding where to start.

EUR 890
  • A review of your repository, your CI and how the team uses AI today
  • A written report: where you stand, and the risks around secrets, permissions and review
  • A rollout plan in steps your team can take
  • A fixed quote for the Harness Build

100% upfront · credited if you order a build within 30 days

Step 3

Care Plan

For: Teams that want the harness kept current.

SEUR 450 / mo
MEUR 850 / mo
  • Up to 3 hours a month on S, up to 6 hours a month on M
  • New skills as your work changes
  • Updates when models and tools change
  • Reviews of the harness, and questions answered

Billed monthly in advance · unused time does not carry over

Not ready for a full build? Start with a half-day remote workshop for up to 12 people: how coding agents work, where they fail, and how a team sets them up safely.Team WorkshopEUR 790

Prices excl. VAT. The assessment fee is credited if you order a build within 30 days.

How an engagement runs

From the first look to a harness that stays current

  1. Assessment

    I review the repository and how the team works with AI now.

    100% upfront
  2. Proposal

    The report ends with a fixed quote for the build.

    Fixed price
  3. Build

    Instructions, skills, hooks and guardrails, built and tested in your repository.

    50% at start
  4. Handover workshop

    The team learns the setup on its own code.

    50% at handover
  5. Care

    New skills and updates as the models and the team change.

    Monthly, in advance

Invoices are due in 8 days. Prices excl. VAT. I take on work in the order it is confirmed, and tell you up front when I can start.

How I work

I run my own development on a harness like this

The numbers come from my own projects, with the date and method beside each one.

tickets closed by merged pull requests, Aug 28 – Oct 27 2026
584
median model cost per ticket at API list prices (implementation + review + fixes), Oct 2026
$15
destructive, history-rewriting or credential commands blocked by guards
108

Measured on my own projects, Aug – Oct 2026. Method on the case study page.

Read the case study

About

I'm László Kovács.

László Kovács

I'm a software architect and backend tech lead with 15 years of building platforms and leading engineering teams of up to 13. Since coding agents became usable for real work, I have done my own development through them, on the setup this site describes, and I build the same kind of setup for other teams. I work remotely from Szeged, Hungary, in the EU time zone, and I take on a small number of teams at a time.

FAQ

Questions teams ask first

Do we already need Claude Code?

No. If the team is just starting, the assessment covers which tool and plan fit you, and the build sets it up. The harness is built for Claude Code; if your team is set on another agent, tell me in the form and I'll say honestly whether I'm the right fit.

Do you need access to our code?

Yes, read access to the repository for the assessment, and write access to a branch for the build — everything I change arrives as pull requests your team reviews and merges. I don't need production systems, customer data or your secrets. Access is removed when the engagement ends, or kept for the Care Plan if you choose one.

What about secrets and NDAs?

I sign your NDA before the assessment starts. Keeping secrets away from the agent is part of every build: the harness limits what the agent can read, blocks it from printing or committing credentials, and scans outgoing code for secrets before it leaves the machine.

How long does a build take?

It depends on the repository and the team, and the assessment tells you. I work on a few engagements at a time, so I give you a start date before you order, not a promise I'd have to break.

What happens after the handover?

The harness is yours: it lives in your repository and your team can change it like any other code. The Care Plan is optional — it is there if you want someone keeping it current as models and tools change.

Contact

Tell me where your team is

A few questions by email first, then a short call if it's a fit. I reply within 2 business days.

Where are you now?

I use these details only to reply to you. See Privacy.

Prefer email? [email protected]