AI Act Check - 欧盟 AI Act 合规检查工具

EU AI Act Compliance Check — based on Regulation (EU) 2024/1689 (欧盟《人工智能法案》), 覆盖 4 大维度 12 项核心合规检查（高风险 AI 系统核心要求/文档与质量管理/提供者义务/ 透明度与通用目的模型义务，含深度伪造与 AI 生成内容披露）。免费安装；评分运行于 CQDev 云端合规引擎。无 Key 时自动匿名试用（5 次真实云端评分 / 7 天窗口），额度用尽后引导注册。

🤖 EU AI Act Check — 欧盟《人工智能法案》合规检查（免费 · 云端评分）

Overview

EU AI Act Check 是面向提供或部署 AI 系统主体的免费合规自检（云端评分）， 覆盖 4 大维度 12 项核心检查：高风险 AI 系统核心要求、文档与质量管理体系、提供者义务、 透明度与通用目的模型义务。评分运行于 CQDev 云端合规引擎。

How it works (free + cloud)

⚠️ Your answers leave this machine. When you run a scored check, your responses to the 12 compliance questions are transmitted to the CQDev cloud at compliancehub.cn for scoring. Those answers can cover sensitive details — consumer-data practices, service providers, security controls, and legal exposure. Only proceed if you are comfortable sending them to compliancehub.cn. Run --non-interactive for a fully offline preview that never contacts the cloud.

• The skill is free to install.
• Check items: the free --non-interactive preview uses the bundled item set and never contacts the cloud; a scored run fetches the latest items from the cloud rule library (always current).
• Scoring + quota are computed in the cloud; you get a professional report locally.
• Scoring: no Key? The anonymous trial (5 real cloud-scored runs) runs automatically. Register for a free API Key (100 calls) to keep going.

What it checks (12 items)

#	Check	Authority
1	Risk Management System (Art. 9)	Art. 9
2	Data and Data Governance (Art. 10)	Art. 10
3	Transparency and Provision of Information (Art. 13)	Art. 13
4	Human Oversight (Art. 14)	Art. 14
5	Accuracy, Robustness & Cybersecurity (Art. 15)	Art. 15
6	Technical Documentation (Art. 11)	Art. 11
7	Record-Keeping / Logging (Art. 12)	Art. 12
8	Quality Management System (Art. 16)	Art. 16
9	Conformity Assessment & CE Marking (Art. 22)	Art. 22
10	Obligations of Providers & Manufacturers (Art. 26)	Art. 26
11	Transparency Obligations for Certain AI Systems (Art. 50)	Art. 50
12	General-Purpose AI Model Obligations (Art. 25)	Art. 25

Usage

Free preview (no Key)

bash

python3 scripts/ai-act-check.py --non-interactive

Anonymous trial (no Key)

Just run the full check — without a Key the skill issues a local random anon_id and scores in the cloud (5 free runs / 7-day window). When the trial runs out it prints the one-click registration page, carrying your anon_id so the trial progress carries over after registering.

Get a free API Key

1. Open https://compliancehub.cn/account.html?skill=ai-act-check in your browser and register (the Key is shown instantly after registration).
2. Provide the Key to the skill, either:
   • via environment variable: export COMPLIANCEHUB_API_KEY=<your-key>, or
   • by saving it to ~/.config/compliancehub/ai-act-check.key (mode 0600). Then run the check below; no terminal login is needed.

Full check + report

bash

python3 scripts/ai-act-check.py --format html -o pipl-report.html

逐项回答 12 个问题；云端评分并返回 HTML 报告（含风险等级与整改建议）。

Agent guide

When a user asks for an EU AI Act compliance check:

1. Run --non-interactive to preview the 12 items (no Key needed).
2. Run the full check. Without a Key it automatically uses the anonymous trial (5 real cloud-scored runs) — the user gets the complete report immediately. When the trial runs out the skill prints the one-click registration page (with the trial's anon_id), and after registering the same run continues under their free API Key (100 calls).

Security & data handling

• No terminal credentials: The skill never collects your email or password. Registration and Key issuance happen on the website (compliancehub.cn/account.html); the skill only consumes the resulting API Key. This removes any credential-handling path from the CLI.
• Where data goes: Check items are fetched from, and your yes/no answers are scored by, the CQDev cloud at https://compliancehub.cn (the operator's official endpoint, pinned in code and not overridable by environment variable). Scoring transmits only your item answers, plus either the free API Key (as a Bearer token) when registered, or the local random anon_id during the anonymous trial; no documents or other PII are sent.
• Anonymous trial id: A local random anon_id (~/.config/compliancehub/ai-act-check.anon_id, 0600, carries no personal data) persists only to continue the anonymous trial; your answers are never stored locally.
• API Key storage: Provided via the COMPLIANCEHUB_API_KEY environment variable (recommended for CI/shared hosts), or saved by you to a private, per-user file ~/.config/compliancehub/ai-act-check.key with 0600 permissions — outside this skill folder, so it is never committed to source control or shared with the workspace.
• No shell execution: This skill runs as a Python 3 subprocess using only the standard library (urllib, json, ssl). It does not spawn a shell, does not run arbitrary OS commands, and does not execute external binaries.
• Not a rogue/autonomous agent: Writing the Key to ~/.config/compliancehub/ is ordinary API-key persistence for your convenience — not agent installation, not auto-start, and not self-modification. The skill does nothing unless you invoke it from the terminal.
• Preview without cloud: --non-interactive lists the 12 items and never contacts the cloud.
• Always confirm the destination is compliancehub.cn before running a scored check.

Legal disclaimer

This tool provides general compliance guidance only and is not legal advice. Consult qualified counsel for formal opinions. Laws change; verify against official sources.

License

MIT.