Custom builds by Wondel.ai

We build the skills you already use. Now we'll build yours.

Custom Claude skills, subagents & MCP integrations — from the team behind the open-source wondelai/skills. Shipped to production, not demoed.

1,451 stars on GitHub

For teams who want senior-level Claude work shipped — not another impressive demo.

SKILL.md 
---
name: acme-support-triage
description: Triage inbound support tickets the
  way our senior team does — classify, draft, escalate.
---

## When to use
Every new ticket that lands in the queue.

## Steps
1. Classify: billing · bug · feature · churn-risk
2. Pull account history via the Zendesk MCP.
3. Draft a reply in our voice, and cite the doc.
4. Escalate churn-risk to #cs-leads.
The problem

Impressive in the demo.
Gone by the pilot.

Demos that die in pilot

The proof-of-concept dazzles in the meeting, then stalls before it ever reaches production. Impressive is not the same as shipped.

No internal skills library

Every engineer prompts from scratch, pasting the same 200 words of context each time. Nothing compounds, nothing is reusable.

Agents no one can trust

Without evals, no one can say whether the agent is right — so it stays a toy, never wired into the work that actually matters.

Most AI projects die in the gap between impressive and shipped.

What we build

Three layers of one system.

Skills, agents and the plumbing that connects them — built against evals, shipped into your stack.

Custom Skills

Your team's know-how, packaged as skills Claude loads every session.

  • A portable SKILL.md with scripts and resources
  • Progressive disclosure that keeps context lean
  • Versioned in your repo or shipped as a plugin

Custom Agents & subagents

Reviewers, debuggers, data-scientists — agents scoped to do one job well.

  • Subagents in Claude Code, or standalone on the Claude Agent SDK
  • Least-privilege tools and clear degrees of freedom
  • Model routing tuned for cost and quality

MCP & internal libraries

Secure connectors into your systems, plus a library that compounds.

  • Private MCP servers onto your proprietary tools and data
  • An internal .claude/ setup: skills, subagents, hooks, MCP
  • An eval suite that stays a quality asset

MCP is the hands, Skills are the know-how, subagents are the orchestration — a serious build usually needs all three.

The proof is public

You've already used our work.

wondelai/skills on GitHub Built on the agentskills.io open standard

Founded by Michał Jaskólski — built & sold two companies, shipped consumer products to 4M+ users a month. This studio ships the open skills you can already read, star and run.

Refactoring UI iOS HIG Design UX Heuristics Web Typography Design of Everyday Things Top Design StoryBrand Messaging Made to Stick Contagious 1-Page Marketing Plan Scorecard Marketing Obviously Awesome $100M Offers Negotiation Influence Psychology Predictable Revenue CRO Methodology Jobs to Be Done Hooked UX Improve Retention Lean Startup Design Sprint Crossing the Chasm Blue Ocean Strategy Traction EOS Drive Motivation Clean Code Pragmatic Programmer Refactoring Software Design Clean Architecture Domain-Driven Design Data-Intensive Apps System Design Release It! High Perf Browser Continuous Discovery Inspired The Mom Test The 37signals Way Lean UX Microinteractions Steve Jobs Design Review Cold Start Problem Good Strategy Bad Strategy High Output Management Team Topologies Lean Analytics Monetizing Innovation Legacy Code + yours Refactoring UI iOS HIG Design UX Heuristics Web Typography Design of Everyday Things Top Design StoryBrand Messaging Made to Stick Contagious 1-Page Marketing Plan Scorecard Marketing Obviously Awesome $100M Offers Negotiation Influence Psychology Predictable Revenue CRO Methodology Jobs to Be Done Hooked UX Improve Retention Lean Startup Design Sprint Crossing the Chasm Blue Ocean Strategy Traction EOS Drive Motivation Clean Code Pragmatic Programmer Refactoring Software Design Clean Architecture Domain-Driven Design Data-Intensive Apps System Design Release It! High Perf Browser Continuous Discovery Inspired The Mom Test The 37signals Way Lean UX Microinteractions Steve Jobs Design Review Cold Start Problem Good Strategy Bad Strategy High Output Management Team Topologies Lean Analytics Monetizing Innovation Legacy Code + yours
50
Open-source skills
8
Domains covered
4M+
Users shipped
2
Companies exited
How we work

Eval-driven, from day one.

01

Discovery

We map the time-consuming, repeatable workflows — the tasks where your team pastes the same context every single time.

02

Baseline & evals

Before a line of code, we run Claude without the skill, document where it fails, and turn those failures into evals. Evals become the source of truth.

03

Design & build

We pick the right primitive — Skill, subagent, MCP, hook or Agent-SDK app — and build against the evals, not vibes.

04

Ship to production

Version-controlled, integrated with your tools, and deployed where your team already works — not left behind in a notebook.

05

Iterate & maintain

We watch real usage, expand the evals, and keep everything current as the models and your needs move.

Evals are the deliverable, not an afterthought.

Engagements

Start small. Scale what works.

Skill / Agent Sprint

from $3K
1–2 weeks

A fixed-scope paid POC that de-risks the jump from pilot to production.

  • One skill or agent, shipped
  • Baseline + eval suite
  • Handover and docs
Book a scoping call
Most chosen

Custom Build

from $15K
fixed scope

A production build — skills, agents and the MCP plumbing that connects them.

  • Multi-skill / multi-agent system
  • Private MCP integrations
  • Shipped to production, in your stack
  • Team enablement
Book a scoping call

Skills-Library-as-a-Service

from $3K/mo
retainer

An evolving internal library of skills and agents that compounds every month.

  • Ongoing skill & agent development
  • Growing eval coverage
  • Model-upgrade maintenance
  • Priority support
Book a scoping call

Shipped to production, or you don't pay the final milestone.

Straight answers

The questions you're actually asking.

Is this just prompting?

No. The work is evals and context engineering — defining what “good” means, then designing skills, tool scoping and model routing to hit it repeatably. Prompting is the easy 10%.

Who maintains it after launch?

You can, or we do. The Library retainer keeps your evals green, expands coverage, and handles model upgrades so nothing quietly drifts.

Is our data and security safe?

We treat a skill like installing software: least-privilege subagents, MCP servers scoped to exactly what they need, and secrets kept off surfaces that never require them.

Are we locked into Claude?

Skills are built on the open agentskills.io standard, so the know-how stays portable to other compatible agents. You own the artifacts either way.

Who's behind this

A studio, with a founder-operator who has actually shipped — not a faceless agency, and not a $50/hr freelancer.

Michał Jaskólski Founder, Wondel.ai

Michał has built and sold two companies — Nokaut.pl (~$22M IPO) and Morizon.pl (~$20M M&A) — and shipped consumer products to 4M+ users a month as CPO of Grupa Morizon, now part of Ringier Axel Springer. Today he builds Wondel.ai and Czat.ai, and the open-source skills this studio is built on.

Let's talk

Let's build yours.

A quick 15-minute call: bring a workflow, leave with a clear plan and a fixed-scope first sprint.

Book a quick call

Prefer email? [email protected] · or start with the free skills.