This guide explains how to write a new skill that fits the repository conventions.

Skill anatomy

Each skill lives in <area>/skills/<skill-name>/SKILL.md and follows this structure:

---
name: skill-name
description: "Eine prägnante Beschreibung..."
language: de
agents:
  researcher: ../../agents/researcher.md
  drafter: ../../agents/drafter.md
  reviewer: ../../agents/reviewer.md
provider_variants:
  - claude
  - gemini
  - openai
test: ./test.md
---

# /area:skill-name

## Zweck
...

## Eingaben
...

## Sub-Agent-Architektur
...

## Ablauf
...

## Quellen und Zitierweise
...

## Ausgabeformat
...

## Risiken / typische Fehler
...

Required sections

  1. Zweck — what the skill does and when to use it.
  2. Eingaben — what data the user must provide.
  3. Sub-Agent-Architektur — how researcher / drafter / reviewer collaborate.
  4. Ablauf — the step-by-step process.
  5. Quellen und Zitierweise — statutes, commentary, case-law (citations linked).
  6. Ausgabeformat — the deliverable shape (letter, memo, table, etc.).
  7. Risiken / typische Fehler — known pitfalls.

Citation discipline

- (no marker, with URL) — verified against an official source. - [unverifiziert – prüfen] — model knowledge, not externally checked. The user must verify. - [generiert] — forbidden in client-facing output.

See references/zitierweise.md for the binding citation conventions.

Methodology

Test file

Every skill ships with a test.md that captures at least one full input/output trace. The test file is referenced from the frontmatter (test: ./test.md).

Submit

Open a PR against main. The CI pipeline will run the smoke tests and check the frontmatter schema.