Install
$ npx skills add Castaldo-Solutions/process-builderREADME
# GitHub Repository: Castaldo-Solutions/process-builder
**URL:** https://github.com/Castaldo-Solutions/process-builder
**Author:** Castaldo-Solutions
**Description:** Open-source AI skill that turns a process interview into a professional BPMN swimlane diagram (.drawio). AS-IS with pain points, TO-BE with Quick Win / Full Automation.
**Homepage:** https://www.castaldosolutions.it/articles/blog/mappare-processi-aziendali-ai-swimlane
**Language:** Python
## Stats
- Stars: 9
- Forks: 0
- Open Issues: 0
- Commits: 4
- Created: 2026-07-04T15:12:44Z
- Updated: 2026-08-22T16:02:56Z
- Pushed: 2026-08-21T20:10:43Z
## README
# Process Builder: AI Skill for Business Process Diagrams
[](LICENSE)
[](scripts/generate_swimlane.py)
[](scripts/generate_swimlane.py)
[](../../pulls)
An open-source [Claude Skill](https://docs.claude.com/en/docs/agents-and-tools/agent-skills) that turns a **conversation about a business process** into a professional **swimlane diagram** in [draw.io / diagrams.net](https://www.diagrams.net/) format, with hybrid **BPMN 2.0** notation, curated palette, pain points mapped where they hurt.

*The diagram above was generated from a single JSON file: no manual drawing, no XML editing.*
## What it does
- **AS-IS mode**: the agent interviews the user about how a process works today (who does what, when, with which tool, where it hurts) and produces a diagram with actors as lanes, BPMN gateways for decisions, tool annotations, and **pain points** highlighted outside the flow.
- **TO-BE mode**: starting from an AS-IS, it proposes an optimized process in two variants: **Quick Win** (2–4 weeks, light automation) and **Full Automation** (deep integrations), with automated/new tasks visually marked.
- **Interview-first, always**: the skill enforces a hard rule: no diagram until the agent knows who does what, in which order, with which tool, and which decisions exist and what they depend on. No plausible-but-invented processes.
> **Note:** the skill content (prompts, interview guides) is in **Italian**, as it was designed for process-mapping interviews with Italian-speaking teams. The generator script and its JSON schema are language-agnostic.
## From a tiny JSON…
```json
{
"phases": [{
"name": "PROCESSO - RICHIESTA FERIE",
"actors": ["Dipendente", "Responsabile", "HR"],
"elements": [
{"id": "s1", "type": "start", "actor": "Dipendente"},
{"id": "a1", "type": "action", "actor": "Dipendente", "label": "Compilare modulo ferie", "tool": "Excel"},
{"id": "g1", "type": "gateway_xor", "actor": "Responsabile", "label": "Approvata?"},
{"id": "pp1","type": "pain_point", "target": "a1", "label": "Il dipendente non conosce lo stato della richiesta"}
],
"connections": [
{"from": "s1", "to": "a1"},
{"from": "a1", "to": "g1"}
]
}]
}
```
## …to a boardroom-ready diagram

## How it works
The agent never writes draw.io XML by hand. It produces a small JSON description of the process and runs:
```bash
python scripts/generate_swimlane.py process.json process.drawio
```
The script validates first (actionable errors for duplicate ids, orphan connections, unknown actors), then handles the whole layout: columns, lanes, phases, multi-row pain points, and **collision-free arrow routing** (gateway exits are automatically assigned to distinct sides of the diamond, long same-lane jumps hop over intermediate elements).
```bash
python scripts/generate_swimlane.py process.json --validate # check only
python scripts/generate_swimlane.py process.json out.drawio --strict # warnings block too
```
Requirements: **Python ≥ 3.9**, zero third-party dependencies.
## Try it in 30 seconds
```bash
python scripts/generate_swimlane.py examples/richiesta_ferie_as_is.json ferie.drawio
```
Open `ferie.drawio` in [app.diagrams.net](https://app.diagrams.net/) or in draw.io Desktop.
## Using it as a Claude Skill
Copy this repository into a `process-builder` folder inside your Claude skills directory (or grab `process-builder.skill` from the [latest release](../../releases/latest)), then ask something like:
> "Let's map our order-management process: draw the AS-IS."
The agent will interview you, confirm the reconstruction, and hand you the `.drawio` file.
## Repository layout
| Path | Purpose |
|------|---------|
| `SKILL.md` | The skill entry point: workflows, hard rules, quality checklist |
| `references/interview-guide.md` | How to run the AS-IS / TO-BE interview, and when a process deserves multiple phases |
| `references/style-guide.md` | Visual style: shapes, palette, BPMN conventions |
| `references/drawio-xml-guide.md` | JSON input schema for the generator script |
| `scripts/generate_swimlane.py` | JSON → .drawio generator (validation + layout) |
| `examples/` | Ready-to-run example JSONs (single phase, 3 phases with link events, TO-BE with info panels) |
| `evals/` | Evaluation scenarios used to test the skill |
## License
[MIT](LICENSE). Use it, fork it, ship it.
---
Maintained by [**Castaldo Solutions**](https://www.castaldosolutions.it), enterprise-grade technology built for SMEs.
The thinking behind this skill, and what we learned mapping real processes in Italian SMEs: [Mapping business processes with AI](https://www.castaldosolutions.it/articles/en/blog/mappare-processi-aziendali-ai-swimlane).
Information
Repository
Language
Python
Created
2026/9/5
Updated
2026/9/5