Guide · Agent skill
How to use the AI System Inventory (EU AI Act) agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The ai-inventory skill is an agent skill that turns Claude Code, or a compatible agent, into a record-keeper for the AI systems your organization uses under the EU AI Act. It adds systems through a short interview, lists them in a table with counts by risk tier, and walks through the role and risk tier for each one. This guide explains what an agent skill is, how to install this one, how to trigger it with plain requests, and what it needs before you start.
AI System Inventory (EU AI Act)anthropics/claude-for-legal · Apache-2.0 Download the skill Quick start
- 1Download and unzip. Download the skill ZIP from the directory page and unzip it. Place the folder in ~/.claude/skills/ or in your project's .claude/skills/ folder, or upload it under Settings, then Capabilities, then Skills in the Claude apps.
- 2Run the setup interview. Run /ai-governance-legal:cold-start-interview so that the CLAUDE.md config file in the ai-governance-legal folder is filled in.
- 3Add your first system. Ask to add an AI system or run /ai-inventory add. Answer the questions for name, owner, description, status and EU nexus, one at a time.
- 4Run the classification walk-through. When the skill offers it, go through the role and then the tier questions. Record the role_basis and tier_basis that come out of the conversation.
- 5Review the list. Run /ai-inventory list to see the table, counts by tier and any systems flagged for review within 30 days.
What is an agent skill, and what does this one do?
An agent skill is a folder containing a SKILL.md file. The file tells the agent when to use the skill and how to carry out a specific task. This skill is for the EU AI Act. It keeps a structured inventory of AI systems, with one record per system.
The core idea is that role and risk tier are assessed per system, not per company. One organization can be a provider of one system, a deployer of another, and an importer of a third. The skill records each assessment next to the system it applies to. It does not derive legal obligations from a lookup table. Obligations are worked out in conversation, and the skill flags where a lawyer must verify the article mapping against the current AI Act text.
- Add a system through a short interview, one field at a time.
- List the inventory as a table with counts by risk tier.
- View, edit or classify an individual record.
- Store records in a YAML file on your own machine.
How do you install the ai-inventory skill?
Download the ZIP for the skill from the directory page and unzip it. The skill is a single SKILL.md file. Put the unzipped folder in one of these locations:
For all of your projects, place it in ~/.claude/skills/. For one project only, place it in that project's .claude/skills/ folder. You can also upload the skill in the Claude apps under Settings, then Capabilities, then Skills.
- Personal install: ~/.claude/skills/
- Project install: .claude/skills/ inside your project folder
- Claude apps: Settings → Capabilities → Skills, then upload
Run the setup interview before anything else
The skill reads a config file called CLAUDE.md from the local folder ~/.claude/plugins/config/claude-for-legal/ai-governance-legal/. If the file is missing, or still contains [PLACEHOLDER] markers, the skill sends you to the setup step first. Run the setup command once, before your first inventory action.
The inventory itself lives in ai-systems.yaml in the same folder. If that file does not exist, the skill creates it with an empty systems list when you add the first system.
/ai-governance-legal:cold-start-interviewHow do you trigger the skill with natural requests?
You can use the slash command or plain language. The skill responds to these phrases: ai inventory, add an ai system, what systems do we have, classify this ai system, eu ai act register and ai system registry.
The slash command takes one of five arguments: list (the default when no argument is given), add, edit <id>, classify <id>, and show <id>. Records have IDs in the form sys-001, sys-002 and so on.
- "What systems do we have?" shows the inventory table.
- "Add an AI system" starts the interview.
- "Classify this AI system" starts the role and tier walk-through for an existing record.
- "Show sys-002" displays the full record.
/ai-inventory list
/ai-inventory add
/ai-inventory classify sys-001
/ai-inventory show sys-001What does an inventory record look like?
Each system needs five required fields: name, owner, description, status and EU nexus. Status is one of planned, in_development, in_production or deprecated. EU nexus asks whether the system is deployed in the EU or EEA, offered to users there, or used to produce outputs that affect people there. If any of these is true, EU AI Act analysis applies.
The classification walk-through adds a role, a role basis in one sentence, a tier and a tier basis. The skill tags both bases for verification. The list view also shows an owner and a next review date, and the skill counts systems flagged for review within 30 days.
- Role options: provider, deployer, importer, distributor, authorized representative, product manufacturer.
- Tier options: prohibited, high_risk, limited, minimal, GPAI, GPAI + systemic risk.
- If you substantially modify a vendor system, the skill flags that you may have become a provider of the modified system.
What should you check before relying on the output?
The skill is a structured record and a prompt for analysis. It is not legal advice. Its own guidance says the article mappings are complex and must be verified by a lawyer against the current AI Act text. Treat every role and tier basis as a draft until counsel has reviewed it.
The skill runs no scripts and makes no network calls. Its only file access is to the config folder and the inventory YAML file in that folder.
- Verify every role and tier basis with a lawyer.
- Keep the inventory file backed up, since it lives on your machine.
- Expect the skill to refuse to derive obligations from a table.
What people use it for
- Keeping a register of company AI tools that touch people in the EU, with an owner for each.
- Recording that the same company is a deployer of a vendor's hiring tool and a provider of an in-house model.
- Preparing a per-system record of role and risk tier, with reasons, for a governance or compliance review.
- Tracking which of several AI tools have an EU nexus and which do not.
Frequently asked questions
Is the ai-inventory skill free to use?
The source does not state pricing or licence terms beyond the Apache-2.0 licence listed for the item. Check the licence file in the repository for the terms that apply to you.
What do I need to run the ai-inventory skill?
You need Claude Code or a compatible agent with access to the local ~/.claude/plugins/config/claude-for-legal/ai-governance-legal/ folder. You also need to complete the setup interview, which creates the CLAUDE.md config file. The skill runs no scripts and uses no network.
Where does the skill store my AI system data?
It stores records in ai-systems.yaml inside the ai-governance-legal config folder on your own machine. The skill does not send the file anywhere.
Does the skill tell me my legal obligations under the EU AI Act?
No. It walks through role and risk tier questions and records the answers, but it does not use a lookup table for obligations. Obligations are discussed in conversation, and a lawyer must verify the article mapping against the current AI Act text.
Why does the skill say nothing works until I run cold-start-interview?
The skill reads CLAUDE.md first. If that file is missing or still has [PLACEHOLDER] markers, the skill directs you to /ai-governance-legal:cold-start-interview before it continues.
Is this skill useful for someone with no EU connection?
The EU-related questions do not apply to individuals with no EU nexus. The skill is aimed at organizations that use AI tools which affect people in the EU or EEA.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use this skill if your in-house legal or governance team needs a structured EU AI Act register of AI systems, with a role and risk tier recorded for each one. Pick a different tool if you need automatic obligation mapping or final legal advice, since the skill deliberately does neither and expects a lawyer to verify its output.
Download the skill