LIVE · v1.2 STABLE
BUILD 2026.05.27
ART. 6
AGPL-3.0 · 10 EVAL CASES · 0 ASSERTIONS

IS YOUR
SYSTEM
high-RISK
UNDER ART. 6.

Depth classification of an AI system against Art. 6 of the EU AI Act, grounded in the Commission’s draft Art. 6(5) guidelines. Annex I safety-component analysis, Annex III use-case mapping, Art. 6(3) exception filter, profiling re-exception, Art. 25 substantial-modification trap, and an AI-Omnibus-corrected effective-date verdict — in one decision tree.

AI ACT ART. 6   ·   ART. 6(1)   ·   ART. 6(2)   ·   ART. 6(3)   ·   ART. 6(4)   ·   ANNEX I   ·   ANNEX III   ·   ART. 25   ·   ART. 3(14)   ·   REG. 2024/1689   ·   DECISION 768/2008/EC   ·   AI OMNIBUS 2026   ·   AI ACT ART. 6   ·   ART. 6(1)   ·   ART. 6(2)   ·   ART. 6(3)   ·   ART. 6(4)   ·   ANNEX I   ·   ANNEX III   ·   ART. 25   ·   ART. 3(14)   ·   REG. 2024/1689   ·   DECISION 768/2008/EC   ·   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
ANNEX I SAFETY COMPONENT
Art. 3(14) two-prong test: safety function (intent, ¶¶37) OR failure/malfunction (consequences, ¶¶38–43). Screens against Union harmonisation legislation — machinery, lifts, MDR, IVDR, automotive.
02
ANNEX III USE-CASE MAP
Cross-checks the intended purpose against all eight Annex III areas: biometrics, critical infrastructure, education, employment, essential services, law enforcement, migration, justice/democracy. Captures the specific sub-point (e.g. Nr. 4(a)).
03
ART. 6(3) EXCEPTION FILTER
Tests the four narrow limbs (a) procedural / (b) improvement / (c) deviation detection / (d) preparatory. Profiling under Art. 4(4) GDPR excludes the exception and re-confirms high-risk.
04
CONFORMITY-ASSESSMENT TRIGGER
Maps the Annex I legal act to its Decision 768/2008/EC module. Modules B–H1 invoke a notified body; Module A conditioned on harmonised standards stays high-risk (¶57).
05
ART. 25 MODIFICATION TRAP
Flags the quasi-provider trap: rebranding, substantial modification, or repurposing a non-high-risk system (incl. GPAI) into a high-risk one under Art. 25(1)(a–c).
06
AI-OMNIBUS DATE CORRECTION
Applies the postponed effective dates: Annex III to 2 Dec 2027, Annex I to 2 Aug 2028, Art. 111(2) legacy cut-off to 2 Dec 2027.
§ 02 · WORKFLOW

FROM INPUT
to ARTEFACT.

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

01
Disclaimer & inputs
Non-blocking disclaimer. Gather system description, intended-purpose evidence, deployment context, and modification status before proceeding.
02
Step 1 — AI system gating
Confirm the Art. 3(1) AI system definition. Pure rule-based software with no inference → not in AI Act scope, terminate.
03
Step 2 — Intended-purpose framing
Document the provider's stated purpose (Art. 3(12); ¶¶10–13). Apply the GPAI / multi-purpose trap: broad positioning deems high-risk uses included.
04
Step 3 — Art. 6(1) / Annex I branch
Is the system a regulated product (¶30) or a safety component (¶31)? Run the Art. 3(14) two-prong test — safety function OR failure-based.
05
Step 3c — Conformity assessment
Look up the required module. Notified-body modules (B–H1) or harmonised-standards-conditioned Module A → high-risk under Art. 6(1).
06
Step 3d — Section A vs B
Record the Annex I citation and whether the legal act is Section A (full Chapter III) or Section B (Art. 6(1), 102–109, 112 only, per Art. 2(2)).
07
Step 4 — Art. 6(2) / Annex III branch
Map the intended purpose to the eight Annex III areas. Capture the matching sub-point for each hit.
08
Step 4b — Art. 6(3) exception
If a use case matches, test the four narrow limbs. An exception means not high-risk — subject to the profiling check.
09
Step 4c — Profiling re-exception
If the system profiles natural persons (Art. 4(4) GDPR), the Art. 6(3) exception is excluded and the system IS high-risk.
10
Step 4d — Exception documentation duty
If the exception holds: not high-risk, but document the reasoning (Art. 6(4)) and register in the EU database (Art. 49(2)).
11
Step 5 — Art. 25 trap
If the user modifies, fine-tunes, or rebrands a third-party system, flag Art. 25(1) quasi-provider status.
12
Verdict & artefacts
Emit the structured decision block, the practitioner memo, and the JSON interchange artefact with citations and effective date.
§ 03 · MODES

TWO PATHS.
One OUTCOME.

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

ANNEX I
Product / safety-component track. Art. 6(1) via Union harmonisation legislation and third-party conformity assessment. Effective 2 Aug 2028.
ANNEX III
Use-case track. Art. 6(2) across the eight high-risk areas, filtered by the Art. 6(3) exception and profiling re-exception. Effective 2 Dec 2027.
§ 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-high-risk/
├── evals  # Test cases + assertions
│   └── evals.json  # 10 cases, 0 assertions
├── references  # Reference corpus
│   ├── ai-omnibus-timeline-postponements.md
│   ├── annex-i-section-a-vs-b.md
│   ├── annex-iii-area-1-biometrics.md
│   ├── annex-iii-area-2-critical-infrastructure.md
│   ├── annex-iii-area-3-education.md
│   ├── annex-iii-area-4-employment.md
│   ├── annex-iii-area-5-essential-services.md
│   ├── annex-iii-area-6-law-enforcement.md
│   ├── annex-iii-area-7-migration.md
│   ├── annex-iii-area-8-justice-democracy.md
│   ├── art-25-substantial-modification-flag.md
│   ├── art-6-1-annex-i-guidelines.md
│   ├── art-6-2-annex-iii-guidelines.md
│   ├── art-6-3-exception-decision-tree.md
│   ├── art-6-general-principles.md
│   └── safety-function-checklist.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-high-risk/ folder
  3. Skill auto-triggers on relevant keywords

CLAUDE CODE / MCP

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

WHAT YOU
get BACK.

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

Structured decision block
Terminal verdict: high-risk YES/NO, branch (6(1) Annex I / 6(2) Annex III Nr. X), basis, Art. 6(3) status, Art. 25 flag, effective date, documentation duty.
Practitioner memo
1–2 page Markdown narrative for a DPO / AI compliance officer: system under assessment, verdict and reasoning, Commission-guideline paragraph anchors, routing, open questions.
JSON interchange artefact
Cross-skill interchange JSON: verdict, basis (article/annex/item/section), Art. 6(3) limb, profiling flag, Art. 25 flag, effective date, citations, downstream routing.
§ 07 · EVALS

TESTED
before SHIPPED.

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

10
Test Cases
0
Assertions
100%
Coverage Required
00
Hi, we are a German GmbH building an AI-based computer-vision system that detect...
0 ASSERTS
01
We are a lift manufacturer
0 ASSERTS
02
We sell a connected smart thermostat in the EU
0 ASSERTS
03
Our HR-tech startup sells a CV-screening tool to EU enterprises
0 ASSERTS
04
We deploy a small AI tool that takes recruiter-written feedback notes about cand...
0 ASSERTS
05
We have an email spam filter that uses ML to classify incoming emails as spam or...
0 ASSERTS
06
We are an enterprise customer
0 ASSERTS
07
Our retail analytics company offers an AI camera system installed at store entra...
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. 2024/1689) Art. 6
High-risk classification rules — Art. 6(1) Annex I, Art. 6(2) Annex III, Art. 6(3) exception, Art. 6(4) documentation duty.
Commission draft Art. 6(5) guidelines
General principles, Annex I, and Annex III classification guidelines issued for stakeholder consultation in 2026 (non-binding).
AI Act Annex I & Annex III
Union harmonisation legislation list (Section A/B) and the eight high-risk use-case areas.
Decision 768/2008/EC
Conformity-assessment module reference (Modules A–H1) determining the third-party / notified-body trigger.
AI Act Art. 25
Substantial-modification / quasi-provider trap for parties altering an existing AI system.
AI Omnibus 2026
Effective-date postponements: Annex III → 2 Dec 2027, Annex I → 2 Aug 2028, Art. 111(2) → 2 Dec 2027.
§ 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 verdict.
Every classification cites the Commission-guideline paragraph (¶37 checklist, ¶46 lifts, ¶170 biometrics) and the article behind it.
VERIFIED
Reproducible classification.
Pin a build; recreate the Art. 6 decision tree years later for audit or conformity-assessment defence.
IMMUTABLE
†
Both branches, in order.
Art. 6(1) Annex I is run in full before Art. 6(2) Annex III — a system can be high-risk under both; neither branch is skipped.
ENFORCED
**
EU-native dates.
Built on the AI Omnibus 2026 postponements, not the superseded Art. 113 dates. Effective-date verdicts are current, not retrofitted.
NATIVE
§ 10 · README.MD

ai-act-high-risk

📄 View the interactive skill page →

Depth assessment skill for the EU AI Act high-risk classification under Art. 6 (Regulation (EU) 2024/1689), grounded in the European Commission's draft Art. 6(5) classification guidelines (general principles, Annex I, Annex III) published for stakeholder consultation in 2026.

For the full classification workflow, see SKILL.md. For version history, see CHANGELOG.md.

What this skill does

Given an AI system description and intended-purpose evidence, determines whether the system is high-risk under:

  • Art. 6(1) + Annex I — AI system as product or safety component of a product under Union harmonisation legislation requiring third-party conformity assessment.
  • Art. 6(2) + Annex III — AI system falling into one of eight high-risk use-case areas (biometrics, critical infrastructure, education, employment, essential services, law enforcement, migration, justice/democracy), subject to the Art. 6(3) exception filter and profiling re-exception.

Outputs a structured decision block + practitioner memo + JSON interchange artefact.

Deployment

Place this skill in your Claude Code skill directory (~/.claude/skills/ or workspace-level). Claude will route to it from ai-act-classifier automatically when a high-risk branch is triggered, or invoke it directly on user prompts mentioning Annex I / Annex III / Art. 6 classification.

Disclaimer

This skill is not legal advice. The Commission guidelines it draws on are draft and non-binding (only the CJEU can give authoritative interpretation). Use under qualified counsel.

License

AGPL-3.0 (see repo-root LICENSE file).