How to use the Customize AI Governance Profile agent skill

What is an agent skill and what does customize do?

An agent skill is a folder containing a SKILL.md file that gives an AI agent specific instructions for one task. When a request matches the skill's description, the agent follows those instructions.

The customize skill works on a profile you already have. The agent reads your saved profile, shows you a grouped map of current values, asks what you want to change, explains what the change affects in other skills, and then writes the update to your config.

• Covers sections such as company details, regulatory footprint, risk posture, people, use case registry, AI system inventory, vendor AI governance, AI policy commitments, impact assessment house style, workflow, and integrations.
• Changes one thing at a time and confirms each change before writing it.
• Never deletes a section. Removing a setting sets it to [Not configured] and explains the effect on the plugin.
• Warns when a change would conflict with other settings or switch off a safety guardrail.

What you need before installing

The skill only adjusts an existing profile. You need the claude-for-legal ai-governance-legal plugin installed and a completed cold-start setup that produces a config file at ~/.claude/plugins/config/claude-for-legal/ai-governance-legal/CLAUDE.md and company-profile.md.

The agent also needs read and write access to the local ~/.claude/plugins/config directory. According to the skill's listing, it runs no scripts and uses no network connection.

How do you install the customize skill?

Download the ZIP file from this page and unzip it. The skill's files are SKILL.md. You can install it in one of two ways.

Option 1: copy the unzipped folder into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder to share it with that project.

Option 2: in the Claude apps, open Settings, go to Capabilities, and upload the skill under Skills.

How do you trigger the skill with a natural request?

You can invoke the skill with the slash command /ai-governance-legal:customize, or describe the change in plain language. The skill's description lists phrases such as "change my [thing]", "update my profile", "edit my config", "tune my playbook", and "customize".

You can also pass a section name or a description as an argument, for example a section name from the map.

• "Change my risk posture to conservative."
• "Update my escalation contacts."
• "Edit my config so the EU AI Act is in scope."
• "Customize my vendor AI positions."

What happens during a typical edit?

The agent first checks that setup has been run. If the config is missing or still contains [PLACEHOLDER] values, it tells you to run /ai-governance-legal:cold-start-interview first and stops.

If setup is complete, the agent lists the customizable sections with a one-line summary of each current value. You pick a section or describe the change. The agent shows the current value, asks for the new one, explains what changes downstream, confirms, and writes the change.

Some changes apply to every plugin. Company name, industry, jurisdictions, practice setting, and stage are written to company-profile.md, and the agent notes that the change affects all 12 plugins.

• Risk posture middle to conservative: more use cases are flagged as conditional rather than approved, more AIA follow-ups are surfaced, and more conservative vendor AI redlines are recommended.
• Adding an escalation contact: skills that route escalations, such as /use-case-triage, /vendor-ai-review, and /reg-gap-analysis, include this contact on the relevant risk bands.
• New use case registry entry: /use-case-triage matches against it on its next run. Existing AIAs are not rewritten, so re-run them if you want the new posture reflected.

What are the safety limits and guardrails?

The skill is built to avoid silent changes. It will not delete a setting, and it flags any change that makes the profile internally inconsistent, such as an aggressive risk posture combined with routing every escalation to the GC, or EU AI Act in scope with no systems flagged for the EU. It asks which side you want.

Some guardrails are adjustable, but a few are structural and should stay in place. The skill names three: the [review] flag, source attribution tags on retrieved content, and [verify] tags on cited statutes and regulations. If you ask to remove one, the agent explains what it protects against and asks you to confirm the trade-off.

The skill edits configuration only. It does not assess any AI system and does not give legal advice.

• Do not use it to disable review flags or verification tags for legal judgment.
• Setup must be completed first. The skill does not run the interview for you.

Tips for making changes safely

Make one change per run. The skill is designed to handle several changes sequentially and confirm each one before moving on, so a single change is easier to check.

Read the downstream explanation before confirming. It tells you which skills will behave differently and whether existing outputs will need to be re-run.

• Use the map first to see the current value before choosing what to change.
• After a change that affects existing assessments, re-run the affected reviews if you need the new setting reflected in them.
• Run /ai-governance-legal:customize again at any time to make further edits.