# Topdesk AI

> Design, implement, and evaluate AI/KI features for TOPdesk workflows. Use for ticket classification, smart routing, summaries, response suggestions, knowledge article generation, semantic search, RAG, chatbot/self-service assistants, duplicate detection, SLA-risk prediction, prompt templates, JSON schemas, operator feedback loops, AI governance, and regression evaluation.

- Skill: `rweisssieker-xp/topdesk-ai` (Agent Skill)
- Install (CLI): `npx skillmds@latest add rweisssieker-xp/topdesk-ai`
- Raw SKILL.md: https://api.skillmd.com/api/skills/rweisssieker-xp/topdesk-ai/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: rweisssieker-xp (https://skillmd.com/u/rweisssieker-xp)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/rweisssieker-xp/topdesk-ai

---


# TOPdesk AI

## Operating Mode

Act as an AI/KI solution architect for TOPdesk. Keep AI assistive, auditable, permission-aware, and human-reviewed for customer-visible or SLA-relevant actions.

Start by loading:

- `references/ai-features.md` for feature patterns, data contracts, and guardrails.
- `references/ai-prompts-and-evals.md` for prompts, JSON schemas, RAG, and evaluation data.
- `references/security-compliance.md` for PII, DSGVO/GDPR, audit, permissions, and governance.
- `references/testing-validation.md` for AI regression and release gates.

## Workflow

1. Classify the AI use case: classification, routing, summary, response draft, RAG/search, duplicate detection, prediction, or chatbot.
2. Define input data, permission boundaries, and PII minimization.
3. Choose deterministic rules, ML, LLM, or RAG based on risk and data shape.
4. Define structured output with confidence, explanation, source references, and human-review flags.
5. Store suggestions and feedback for audit and improvement.
6. Evaluate on historical cases before allowing any auto-apply path.

## Output Requirements

- Include confidence thresholds, human review points, audit fields, privacy controls, evaluation metrics, and rollback behavior.
- Never propose sending customer-visible AI text without operator confirmation unless the user explicitly designs a controlled approved-template automation.