LIVE · v1.3 STABLE
BUILD 2026.05.27
PRÜFBERICHT
AGPL-3.0 · 3 EVAL CASES · 0 ASSERTIONS

THE AI ACT
FILE
as A
PRÜFBERICHT.

Consolidates a full AI Act assessment into a signable, version-controlled Dokumentierter Prüfbericht for legal files and audit trails. context-first intake, input validation gate, 9-section report, citation-grounded prose, jurisdiction overlays, and formatted Word export — built on Regulation (EU) 2024/1689.

REG. (EU) 2024/1689   ·   ART. 2   ·   ART. 3(1)   ·   ART. 5   ·   ART. 6(3)   ·   ANNEX I   ·   ANNEX III   ·   ART. 50   ·   C(2025) 924 FINAL   ·   ISO 22989:2022   ·   OECD AI FRAMEWORK   ·   AI OMNIBUS 2026   ·   REG. (EU) 2024/1689   ·   ART. 2   ·   ART. 3(1)   ·   ART. 5   ·   ART. 6(3)   ·   ANNEX I   ·   ANNEX III   ·   ART. 50   ·   C(2025) 924 FINAL   ·   ISO 22989:2022   ·   OECD AI FRAMEWORK   ·   AI OMNIBUS 2026
§ 01 · CAPABILITIES

SIX PRIMITIVES.
One SKILL.

Each capability is documented separately, tested separately, and called by the workflow at the right moment. Compose them or invoke individually.

01
CONTEXT-FIRST INTAKE
Extracts up to 11 fields from prior classifier, roles, and obligations outputs in the same conversation. Falls back to conversational gathering — maximum 2 interaction turns.
02
INPUT VALIDATION GATE
Phase 1.5 cross-checks classification, role, and completeness before drafting. Flags Annex III mismatches, quasi-provider risk, and missing DPIA (Art. 26(9)) or FRIA (Art. 27).
03
NINE-SECTION REPORT
Scope exclusions (Art. 2), AI system determination (Art. 3(1)), role and territorial scope, risk classification (Art. 5 / Annex I / Annex III / Art. 6(3) / GPAI / Art. 50), obligations, flags, conclusion.
04
CITATION INDEX
Legal-citations index keeps every article reference consistent across sections. Each Yes/No determination carries documented reasoning.
05
JURISDICTION OVERLAYS
Member-State-specific compliance checklists and employment-law overlay feed Section 8 recommendations. Compliance timeline reflects AI Omnibus 2026 postponed dates.
06
WORD EXPORT
Optional .docx: cover page, table of contents, heading hierarchy, styled assessment tables, headers/footers, disclaimer. Naming convention AI-Act-[Template]-[System]-[YYYY-MM-DD].
§ 02 · WORKFLOW

FROM INPUT
to ARTEFACT.

Nine structured steps. The human stays accountable; the skill carries the structure, the citations, and the document trail.

01
Disclaimer & context detection
Non-blocking disclaimer. Detect prior AI Act skill outputs in the conversation or request an Assessment Context block.
02
Coverage analysis
Extract up to 11 fields (system, version, provider, technology, deployment, data, org, sector, role, jurisdiction). Confirm extractions, ask only about gaps.
03
Report-specific intake
Always ask client/matter reference, prepared-by, and date of assessment. Unclear fields marked [UNCLEAR — to be confirmed].
04
Input validation (Phase 1.5)
Classification, role, and completeness cross-checks. Flag Annex III mismatches, quasi-provider risk, missing DPIA / FRIA before proceeding.
05
Template selection
Choose Full Assessment Report, Classification Record (Prüfprotokoll), Compliance Register Entry, Management Briefing (Entscheidungsvorlage), or multiple.
06
Report generation
Populate the 9-section structure from references: report-template, legal-citations-index, interpretation-aids, case-studies.
07
Risk classification
Art. 5 prohibited screening, Annex I product safety, Annex III categories, Art. 6(3) exception, GPAI / systemic risk, Art. 50 transparency.
08
Quality check (Phase 3)
Verify citation completeness, documented reasoning, raised flags, GDPR cross-references, follow-up actions, stated limitations.
09
Word export (Phase 4)
Optional. Generate formatted .docx via docx-processing skill. Cover page, TOC, styled tables, disclaimer. Confirm save path.
§ 03 · MODES

FOUR PATHS.
One OUTCOME.

Match the workflow to the situation. The skill router picks automatically; you can override.

FULL ASSESSMENT REPORT
Default. Comprehensive 9-section Prüfbericht for legal files and regulator submissions.
CLASSIFICATION RECORD
Prüfprotokoll — formal audit trail documenting the classification decision and its reasoning.
COMPLIANCE REGISTER ENTRY
Living compliance tracking document for GRC integration.
MANAGEMENT BRIEFING
Entscheidungsvorlage — 2-page decision document for board / C-level, with enforcement exposure context.
§ 04 · ANATOMY

WHAT'S
IN the SKILL.

Single-folder skill. SKILL.md is the runtime spec; references hold the knowledge corpus; evals hold the proof. 

ai-act-report/
├── evals  # Test cases + assertions
│   └── evals.json  # 3 cases, 0 assertions
├── references  # Reference corpus
│   ├── case-studies.md
│   ├── compliance-timeline.md
│   ├── docx-formatting.md
│   ├── interpretation-aids.md
│   ├── jurisdiction-checklists.md
│   ├── legal-citations-index.md
│   ├── output-templates.md
│   └── report-template.md
├── CHANGELOG.md  # Version history
├── README.md  # Deployment guide
└── SKILL.md  # Main skill instructions
§ 05 · DEPLOYMENT

INSTALL
and INVOKE.

Two deployment surfaces. The skill auto-triggers on relevant keywords once installed.

CLAUDE.AI USER SKILLS

  1. Settings → Profile → Custom Skills
  2. Upload the entire ai-act-report/ folder
  3. Skill auto-triggers on relevant keywords

CLAUDE CODE / MCP

  1. Copy folder to your skills directory:
cp -r ai-act-report/ \
   ~/.claude/skills/user/
§ 06 · OUTPUTS

WHAT YOU
get BACK.

Every output is documented, version-pinned, and traceable to its source citation.

Full Assessment Report
9-section Dokumentierter Prüfbericht: introduction, system description, scope exclusions, scope of application, intended purpose, risk classification, obligations, flags, conclusion.
Classification Record (Prüfprotokoll)
Formal audit trail for the classification decision — the documented basis required before placing an Annex III system on market (Art. 6(4)).
Compliance Register Entry
Living compliance tracking document structured for GRC integration.
Management Briefing (Entscheidungsvorlage)
2-page decision document for board / C-level, including enforcement exposure (up to €35M / 7% turnover).
Word document (.docx)
Formatted export: cover page, table of contents, heading hierarchy, styled assessment tables, headers/footers, final disclaimer.
§ 07 · EVALS

TESTED
before SHIPPED.

Every release runs against a fixed test suite. Assertions check numeric consistency, citation accuracy, and decision-tree branches.

03
Test Cases
0
Assertions
100%
Coverage Required
00
Please generate a full AI Act Examination Report
0 ASSERTS
01
I'd like an AI Act Examination Report for the system described below
0 ASSERTS
02
Quick request — instead of the full Assessment Report, please produce a Manageme...
0 ASSERTS
§ 08 · REGULATORY BASIS

WHAT IT
cites.

Every legal verdict resolves to one of these instruments. No invented articles, no synthetic recitals.

EU AI Act (Reg. (EU) 2024/1689)
Primary instrument — all articles, recitals, and annexes.
Art. 2 / 3 / 5 / 6 / 50–55
Citation anchors for scope, AI system definition, prohibited practices, high-risk classification, and transparency sections.
Commission Guidelines C(2025) 924 final
AI System Definition guidelines (6 Feb 2025); Prohibited AI Practices guidelines (4 Feb 2025).
ISO 22989:2022 · OECD AI Framework
AI concepts, terminology, and AI system definition alignment cited in methodology.
AI Omnibus 2026 timeline
Postponed high-risk dates: Annex III 2 Dec 2027; Annex I 2 Aug 2028.
§ 09 · TRUST

EVERY STEP,
auditable.

The trace is the product. Nothing happens off the record — no hidden tool calls, no silent retrieval, no opaque chain-of-thought.

§
Source-anchored output.
Every classification verdict, role determination, and obligation traces to a cited article from Regulation (EU) 2024/1689.
VERIFIED
Version-controlled file.
Dated, reference-numbered, signable Prüfbericht. Reproduce the same assessment years later for audit or regulatory inquiry.
IMMUTABLE
†
Validation before drafting.
Phase 1.5 gate cross-checks classification, role, and completeness; Phase 3 quality check runs before any document ships.
ENFORCED
**
EU-native.
German legal terms preserved (Prüfbericht, Prüfprotokoll, Entscheidungsvorlage), Member-State jurisdiction overlays, AI Omnibus dates. Not retrofitted.
NATIVE
§ 10 · README.MD

EU AI Act Examination Report Generator — Deployment Guide

📄 View the interactive skill page →

See CHANGELOG.md for version history.

Overview

EU AI Act Examination Report Generator — produces a formal, structured AI Act compliance assessment report suitable for legal files, audit trails, and regulatory inquiries:

  • 9-section report structure — introduction, system description, scope-exclusion check, scope of application, intended purpose, risk classification, applicable obligations, risk flags, conclusion
  • Context-first adaptive intake — consumes prior skill outputs (classifier / roles / obligations / quick) when available
  • Input validation phase — explicit Phase 1.5 to flag missing or inconsistent context before drafting
  • Citation-grounded prose — legal citations index keeps article references consistent
  • Jurisdiction checklists for Member-State-specific compliance items
  • Quality-check phase before delivery (Phase 3)
  • Word document export with formatted output (Phase 4)
  • Output templates for assessment, gap analysis, regulator submission, and internal memo

File Structure

ai-act-report/
├── SKILL.md                              # Main skill instructions (deploy this)
├── CHANGELOG.md                          # Version history
├── evals/
│   └── evals.json                        # Test cases
└── references/
    ├── report-template.md                # 9-section master template
    ├── output-templates.md               # Variant templates (assessment, gap, regulator, memo)
    ├── legal-citations-index.md          # Article-reference index for consistency
    ├── interpretation-aids.md            # Commission/EDPB interpretation hooks
    ├── jurisdiction-checklists.md        # Member-State-specific compliance items
    ├── compliance-timeline.md            # Deadline anchors per tier
    ├── docx-formatting.md                # Word output formatting spec
    └── case-studies.md                   # Worked report examples

Deployment

Claude.ai (User Skills)

  1. Go to Settings → Profile → Custom Skills (or equivalent)
  2. Upload the entire ai-act-report/ folder structure
  3. The skill will auto-trigger on "generate AI Act report", "Prüfbericht", "create compliance assessment", or "export as Word"

Claude Code / Custom MCP Setup

  1. Copy the ai-act-report/ folder to your skills directory: bash cp -r ai-act-report/ /path/to/your/skills/user/ai-act-report/
  2. Ensure the skill is registered in your configuration

Usage

Quick Start

Either start fresh or hand over context from prior skill output:

"Generate a formal AI Act compliance report for our HR screening system. I've already run the classifier (high-risk, Annex III Nr. 4) and the roles skill (we're the deployer). Please assemble the full Prüfbericht and export it as a Word document."

The skill will validate the context, draft the 9-section report, run a quality check, and offer .docx export.

Trigger Phrases

  • "Generate AI Act report" / "Prüfbericht" / "Compliance assessment report"
  • "Document the AI Act analysis" / "Export as Word"
  • "Create a formal AI Act assessment"

Workflow

Phase Description
Phase 1: Input Collection Context-first adaptive intake — consumes prior skill outputs if present
Phase 1.5: Input Validation Explicit gate flagging missing or inconsistent inputs before drafting
Phase 2: Report Generation 9-section template populated with citations and jurisdiction overlays
Phase 3: Quality Check Pre-delivery review for consistency and completeness
Phase 4: Word Export Optional formatted .docx output

Report Structure

Section Content
1. Introduction Purpose, scope of the assessment, methodology note
2. System Description Functionality, deployment context, users
3. Preliminary Check — Scope Exclusions (Art. 2) Military, R&D, personal, ILE, open-source checks
4. Scope of Application Territorial scope (Art. 2), addressee analysis
5. Intended Purpose (Art. 3(12)) Provider-declared intended use
6. Risk Classification Tier verdict with Art. 5 / Annex I / Annex III / GPAI / Art. 50 analysis
7. Applicable Obligations Role × tier obligation map with legal citations
8. Risk Flags & Recommendations Open issues, follow-up actions, monitoring items
9. Conclusion Summary verdict + next steps

Capabilities Summary

Feature Description
9-Section Template Audit-ready structure for legal files and regulator submissions
Context Consumption Reads prior classifier / roles / obligations / quick outputs
Input Validation Gate Phase 1.5 explicitly flags missing/inconsistent inputs
Citation Index Consistent article references across sections
Jurisdiction Overlays Member-State-specific compliance items
Quality Check Phase 3 pre-delivery review
Word Export Formatted .docx output for archiving and distribution
Output Variants Assessment, gap analysis, regulator submission, internal memo

Regulatory Basis

Document Reference
EU AI Act Regulation (EU) 2024/1689
Art. 2, 3, 5, 6, 50–55 Citation anchors for each report section
Commission Guidelines Used in interpretation aids
National implementation Jurisdiction-specific checklist data

License & Disclaimer

This skill produces structured AI Act report templates based on Regulation (EU) 2024/1689. It is not legal advice. Reports should be reviewed and validated by qualified legal counsel before regulatory use.

Licensed under AGPL-3.0 — see LICENSE at the repo root.

Created by Oliver Schmidt-Prietz — OneZero Legal