# Maatwerk AI — free skills for design studios

Four skills for [Claude Code](https://claude.com/claude-code), written for architecture studios,
interior designers and bespoke makers. Each skill is a documented procedure with explicit inputs,
permissions, an approval point and a run record. **AI proposes. AI executes approved tasks.
AI checks. The designer decides.**

| Skill | What it does | Writes to | Approval point |
| --- | --- | --- | --- |
| `brief-register` | Turns a client brief and references into a requirement register with source links, open questions and conflicts | `brief/` | Designer reviews the register before concept work |
| `context-pack` | Builds and maintains a project context pack: sources, approved memory, candidate notes, task context | `context/` | A human promotes candidate notes to approved memory |
| `ifc-check` | Runs deterministic checks on an IFC export and reports checked / failed / not-tested | `checks/` | A named reviewer reads the report before release |
| `sketchup-bounded-op` | Turns one approved modelling instruction into a Ruby script that creates candidate geometry in a new group, then verifies it | `candidates/` | Designer approves the instruction before the script is written, and the result before it is kept |

## Install

```
curl -fsSL https://maatwerk-ai.nl/skills/install.sh | sh
```

Or copy the four folders into `~/.claude/skills/` (every project) or `<project>/.claude/skills/`
(one project). Restart Claude Code. No account, no API key.

## How they fit together

The skills work on their own. They work best inside an **AI Atelier set-up**, which adds the two
things a skill cannot carry by itself:

- **Context engineering** — the sources, approved memory and task context the skill reads from.
  `context-pack` is the first step; the atelier keeps it alive across projects.
- **Harness engineering** — the controls the skill runs within: permissions, approval gates,
  a run ledger and independent checks. The skills write a run record; the atelier enforces it.

Maatwerk AI trains studios on both and builds the atelier around their own tools.
Trainings: https://maatwerk-ai.nl/lab.html · Contact: hello@maatwerk-ai.nl

## Rules every skill follows

1. Never invent a dimension, a source or a decision. Missing stays visible as *missing*.
2. Every claim carries its source: file, revision or date, page or line.
3. Write only inside the skill's own folder (`brief/`, `context/`, `checks/`, `candidates/`). Never modify source documents or existing geometry.
4. Append one entry to `runs/ledger.md` per run (see `_shared/run-ledger.md`).
5. A skill never judges its own output. The check is done by a person or by another run.

## Disclaimer

These skills assist licensed professionals. They replace neither the designer of record nor
the building authority. Output can be wrong; verify it against your sources and standards.

Licence: MIT, © 2026 Demoods B.V. trading as Maatwerk AI (see `LICENSE.md`).