A battle-tested skill collection for Claude Code — built in production inside an autonomous trading system, generalized for any software project.
Core idea: AI writes your code. Governance makes sure you still understand why in six months.
Not prompts. Not templates. Skills — structured workflows that turn Claude Code into a complete development partner: with enforced traceability, machine-level governance, and a real feedback loop between idea and outcome.
Every skill in this repository solves a real problem that emerged while building a live system with 200+ files, 15+ AI agents, and real money on the line — without governance. This is what we learned.
→ Full setup guide (German): HANDBUCH.md
From empty folder to governance-ready project in 5 guided phases — Git hooks, skill set, self-healing monitor and global registry entry included.
Listed in the order you'd typically use them in a development cycle.
| Skill | Command | What it does |
|---|---|---|
| bootstrap | /bootstrap |
Start here. Sets up a new project in 5 phases — CLAUDE.md, Linear integration, Git hooks, full skill set. |
| ideation | /ideation |
Idea → 4-perspective research → Linear issue with acceptance criteria. Prevents gut-feeling decisions. |
| backlog | /backlog |
Sprint planning — which story now, which later, and why. Dependency-aware prioritization. |
| implement | /implement |
8-step protocol: Agent pattern → Spec → Code → Governance validation → Commit. |
| architecture-review | /architecture-review |
Reviews 8 architecture dimensions — risks, tech debt, improvement potential. |
| security-architect | /security-architect |
STRIDE threat modeling, OWASP Top 10:2025, ASVS 5.0. Four modes: Design / Review / Audit / Skill-Scan. |
| research | /research |
2-tier routing: Quick (WebSearch) or Deep (Perplexity sonar + cross-check). |
| sprint-review | /sprint-review |
Quarterly audit: architecture health, tech debt, backlog hygiene. |
| grafana | /grafana |
Grafana Cloud dashboards via MCP — panels, PromQL queries, alert rules directly from Claude Code. |
| cloud-system-engineer | /cloud-system-engineer |
VPS/Docker infrastructure: health checks, firewall, DNS, resources. Usable as a teammate in agent teams. |
| visualize | /visualize |
Generate architecture diagrams in Miro from existing documentation files. |
| skill-creator | /skill-creator |
Create, package and register new skills into the global registry. |
| design-md-generator | /design-md-generator |
Extracts a website's visual design system into a machine-readable DESIGN.md (Google Stitch format) + HTML previews. |
A typical development cycle looks like this:
💡 Idea
└─ /ideation ──→ Linear issue + ACs (4 perspectives, research-backed)
└─ /backlog ──→ Prioritization: which story goes next?
└─ /implement ──→ Spec file → Code → Governance validation → Commit
└─ /architecture-review ──→ Risks? Tech debt?
└─ /sprint-review ──→ Quarterly audit: what worked?
Governance hooks run automatically on every git commit and git push:
spec-gate.sh— blocks commits without a linked spec filedoc-version-sync.sh— blocks pushes when documentation is out of sync
No spec, no commit. That's the difference between a prompt and a governance framework.
| Situation | Recommendation |
|---|---|
| New project, empty folder | → /bootstrap — start here |
| Existing project, needs structure | → HANDBUCH.md §4 — step-by-step retrofit |
| Just one specific skill | → Clone the desired skill folder and install it |
| Want to understand everything first | → HANDBUCH.md — full reference |
- Claude Code (CLI or IDE extension)
- Linear account + API key (issue tracking)
- GitHub repository for your project
- Optional extensions: Grafana Cloud, Miro, Hostinger VPS — skills use what's available
Born from the OpenCLAW Trading System — an autonomous crypto trading bot with 15+ AI agents, 34 self-healing checks, and $200k+ demo capital in production.
Eine battle-tested Skill-Sammlung für Claude Code — entstanden im Produktivbetrieb eines autonomen Trading-Systems, generalisiert für jedes Software-Projekt.
Kernidee: KI schreibt deinen Code. Governance stellt sicher, dass du in 6 Monaten noch weißt warum.
Keine Prompts. Keine Templates. Skills — strukturierte Workflows die Claude Code zu einem vollständigen Entwicklungspartner machen: mit erzwungener Traceability, maschineller Governance und einem echten Feedback-Loop zwischen Idee und Ergebnis.
Jeder Skill in diesem Repository löst ein echtes Problem das entstanden ist, als ein reales System mit 200+ Dateien, 15+ KI-Agents und Echtgeld-Einsatz ohne Governance gebaut wurde. Das hier ist die Lehre daraus.
→ Komplettes Handbuch mit Schritt-für-Schritt-Setup: HANDBUCH.md
Vom leeren Ordner zum governance-ready Projekt in 5 geführten Phasen — Governance-Hooks, Skill-Set, Self-Healing-Monitor und globaler Registry-Eintrag inklusive.
In der Reihenfolge des typischen Einsatzes im Entwicklungs-Workflow.
| Skill | Befehl | Was er tut |
|---|---|---|
| bootstrap | /bootstrap |
Einstieg: Neues Projekt in 5 Phasen aufsetzen — CLAUDE.md, Linear, Git-Hooks, Skill-Set. |
| ideation | /ideation |
Idee → 4-Perspektiven-Research → Linear Issue mit ACs. Verhindert Bauchgefühl-Entscheidungen. |
| backlog | /backlog |
Sprint Planning — welche Story jetzt, welche nach hinten, warum? Abhängigkeiten-aware. |
| implement | /implement |
8-Schritte-Protokoll: Agent-Pattern → Spec → Code → Governance-Validation → Commit. |
| architecture-review | /architecture-review |
Prüft 8 Architektur-Dimensionen — Risiken, Tech Debt, Verbesserungspotential. |
| security-architect | /security-architect |
STRIDE Threat Modeling, OWASP Top 10:2025, ASVS 5.0 — 4 Modi (Design/Review/Audit/Skill-Scan). |
| research | /research |
2-Tier-Routing: Quick (WebSearch) oder Deep (Perplexity sonar + Gegencheck). |
| sprint-review | /sprint-review |
Quartals-Audit: Architektur-Gesundheit, Tech Debt, Backlog-Hygiene. |
| grafana | /grafana |
Grafana Cloud Dashboards via MCP — Panels, PromQL, Alert Rules direkt aus Claude Code. |
| cloud-system-engineer | /cloud-system-engineer |
VPS/Docker-Infrastruktur: Health-Check, Firewall, DNS, Ressourcen. Als Teammate in Agent-Teams einsetzbar. |
| visualize | /visualize |
Architektur-Diagramme in Miro aus bestehenden Doku-Dateien generieren. |
| skill-creator | /skill-creator |
Neue Skills erstellen, paketieren und in die globale Registry einbinden. |
| design-md-generator | /design-md-generator |
Extrahiert das visuelle Design-System einer Website als maschinenlesbare DESIGN.md (Google-Stitch-Format) + HTML-Previews. |
💡 Idee
└─ /ideation ──→ Linear Issue + ACs (4 Perspektiven, Research-backed)
└─ /backlog ──→ Priorisierung: welche Story jetzt?
└─ /implement ──→ Spec-File → Code → Governance-Validation → Commit
└─ /architecture-review ──→ Risiken? Tech Debt?
└─ /sprint-review ──→ Quartals-Audit: Was hat funktioniert?
Governance-Hooks laufen automatisch bei jedem git commit und git push:
spec-gate.sh— blockiert Commits ohne verknüpftes Spec-Filedoc-version-sync.sh— blockiert Pushes wenn Doku-Dateien veraltet sind
| Situation | Empfehlung |
|---|---|
| Neues Projekt, leerer Ordner | → /bootstrap |
| Bestehendes Projekt, Chaos | → HANDBUCH.md §4 |
| Nur einzelne Skills | → Gewünschten Skill-Ordner klonen und installieren |
| Alles verstehen bevor ich anfange | → HANDBUCH.md |
Entstanden aus dem OpenCLAW Trading System — einem autonomen Krypto-Trading-Bot mit 15+ KI-Agents, 34 Self-Healing-Checks und 200k+ Demo-Kapital im Produktivbetrieb.