Skip to content
Get packs

Guide · Agent skill

How to use the AI Usage Policy Starter agent skill (policy-starter)

Updated · written with AI from the project's own documentation and checked against our catalogue data

The AI Usage Policy Starter skill (policy-starter) helps a legal team produce a first-cut draft of an internal AI usage policy. It reads your saved practice profile, asks which sections you want and who will read the policy, searches for published model policies, and drafts the chosen sections with review flags on every judgment call. This guide explains what the skill does, how to install it, how to trigger it, and where its limits are, so you can use it without expecting a finished policy.

AI Usage Policy Starteranthropics/claude-for-legal · Apache-2.0

Quick start

  1. 1Download the ZIP. Download the policy-starter ZIP from this page.
  2. 2Unzip the skill. Unzip into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder.
  3. 3Or upload in the Claude apps. Go to Settings → Capabilities → Skills and upload the ZIP.
  4. 4Populate the practice profile. Complete ~/.claude/plugins/config/claude-for-legal/ai-governance-legal/CLAUDE.md, or run the cold-start interview.
  5. 5Make a request. Ask for an AI usage policy in plain language, or run /ai-governance-legal:policy-starter with an optional scope hint.
  6. 6Answer the interview. Pick the policy sections, then give the audience and deployment context.
  7. 7Review the draft. Work through every [review] flag and verify the cited sources before adoption.

What the policy-starter skill makes the agent do

An agent skill is a folder containing a SKILL.md file with instructions that Claude follows when a request matches the skill's description. policy-starter is designed for teams that have no written AI usage policy yet, or whose older policy does not address newer state AI laws or the EU AI Act.

The skill works in a fixed order. It first reads your practice profile at ~/.claude/plugins/config/claude-for-legal/ai-governance-legal/CLAUDE.md. If that profile is unpopulated, it stops and directs you to /ai-governance-legal:cold-start-interview. It then runs a scope interview before drafting, searches the web for current published model policies, drafts the sections you selected, and closes with a sources block, a reviewer note, an adoption checklist and next steps.

  • Output is labelled DRAFT FOR INTERNAL LEGAL REVIEW — NOT FOR DISTRIBUTION.
  • Every choice point and open question is marked [review].
  • Model policy sources named in the skill include ABA, state bar guidance, ILTA, CLOC, NIST, and public peer-firm or peer-company policies.
  • If the practice profile already contains AI policy commitments, the skill treats the job as an update to the existing policy rather than a new draft.

Install the policy-starter skill

You can install the skill in one of two ways. Download the ZIP file from this page and either unzip it into your personal skills folder or into a project's skills folder, or upload it in the Claude apps.

For a personal install, unzip the folder into ~/.claude/skills/. For a single project, unzip it into the project's .claude/skills/ folder. In the Claude apps, go to Settings → Capabilities → Skills and upload the ZIP.

  • Personal: ~/.claude/skills/
  • Project: .claude/skills/ inside the project folder
  • Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Set up the practice profile first

The skill depends on a practice profile file at ~/.claude/plugins/config/claude-for-legal/ai-governance-legal/CLAUDE.md. The skill reads sections such as Company profile, Use case registry, AI policy commitments, Vendor AI governance, Governance team and escalation, and Who's using this.

If the profile is empty, run the cold-start interview, /ai-governance-legal:cold-start-interview, which the skill points you to. The profile's contents determine whether the draft is treated as new or as an update, and whether the header and framing suit a lawyer or a non-lawyer reader.

  • Company profile: AI role, regulatory footprint, external commitments, practice setting
  • AI policy commitments: if populated, the draft becomes an update to that policy
  • Who's using this: the reader role changes the header and adoption framing
/ai-governance-legal:cold-start-interview

Trigger the skill with natural requests

The skill activates on requests such as the ones below. You do not need to type the slash command. You can also use the command directly and add an optional scope hint in quotes.

The skill's argument hint suggests scope phrases such as firm-wide, legal team only, or update existing.

  • "Draft an AI policy"
  • "We need an AI policy"
  • "Build an AI usage policy"
  • "Our firm needs a GenAI policy"
/ai-governance-legal:policy-starter /ai-governance-legal:policy-starter "we need an AI policy for our 30-lawyer firm" /ai-governance-legal:policy-starter "update our existing policy for the 2026 state AI laws"

Answer the scope interview before drafting

The skill does not start drafting straight away. It asks which policy sections you want, from scope, permitted and prohibited uses, approval and review, disclosure, data handling, training, incidents, enforcement, review cadence and glossary. Its default starter pack for a team with no policy is sections 1, 2, 3, 4, 5 and 9.

It then asks for the audience (all staff, legal team only, attorneys plus staff, or a client-facing version) and the deployment context (law firm, in-house legal, legal aid or clinic, or government). The deployment context decides which model policies the skill searches for.

  • Default starter pack: scope, permitted and prohibited uses, approval and review, disclosure, data handling, review cadence
  • Audience options: all staff, legal team only, attorneys plus staff, client-facing version
  • Deployment options: law firm, in-house legal, legal aid or clinic, government

Review the draft and its flags

The draft is built from the model policies the skill found, adapted to your profile. Read every [review] flag. These mark thresholds, named tools, disclosure triggers and enforcement consequences, which are the points where a person must decide.

Check the cited sources yourself before adoption. The skill's own description says that cited sources and current law should be verified by a human, because web search results cannot be assumed correct.

  • Confirm each cited source says what the draft attributes to it.
  • Decide every [review] item; the skill does not recommend positions on contested questions.
  • Take the draft to the decision-maker, such as the GC, managing partner, board or head of IT.

What people use it for

  • A 30-lawyer firm with no written AI policy that needs a first draft to take to its managing partner.
  • An in-house legal team updating a 2024-era policy to cover state AI laws or the EU AI Act.
  • A legal operations lead preparing a structured set of options for an executive committee or board.

Frequently asked questions

Is the policy-starter skill free to use?

The source material does not state pricing or licence terms for Claude itself. The skill's listed licence is Apache-2.0, so check that licence file for terms that apply to the skill's code and files.

What do I need to run policy-starter?

You need the practice profile at ~/.claude/plugins/config/claude-for-legal/ai-governance-legal/CLAUDE.md, either populated or completed through the cold-start interview. The skill also needs web search access to retrieve current published model policies and guidance. It does not run scripts, but it does use the network.

Where do I install the skill?

Unzip the downloaded ZIP into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. Alternatively, upload the ZIP in the Claude apps under Settings → Capabilities → Skills.

Can the skill produce a final AI policy my company can publish?

No. The skill produces a draft for internal legal review and labels it DRAFT FOR INTERNAL LEGAL REVIEW — NOT FOR DISTRIBUTION. It surfaces choices with [review] flags rather than deciding them, so a human with authority must finalise and approve the policy.

Does the skill use the internet, and is that safe?

Yes, it uses web search to find published model policies such as ABA, state bar, ILTA and CLOC material. The skill runs no scripts. Because search results can be wrong or out of date, verify every cited source and check current law before adopting anything.

How is policy-starter different from writing a policy from a template?

The skill starts from the published model policies it finds for your deployment context and adapts them to your practice profile. It also asks about scope first and marks each judgment call, so you get a structured starting point rather than a generic document.

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 policy-starter if your legal team needs a structured, sourced first draft of an AI usage policy to put in front of decision-makers, and you can spend time checking the flags and sources. Pick something else if you need a policy that is final and approved without attorney review, or if you expect the skill to make firm calls on contested questions, since it surfaces choices rather than deciding them.

Download the skill