Skip to content

Guide · Agent skill

How to use the Policy Redraft Drafter agent skill

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

The policy-redraft skill is an agent skill for Claude that takes a gap in an internal policy, already found by a gap-finding or policy-comparison step, and produces a first-draft, marked-up fix. It is written for in-house legal and compliance staff and policy owners who need a reviewable redline rather than a blank page. By the end of this guide you will know what the skill does, how to install it, how to ask for it in plain language, and what limits to keep in mind.

Policy Redraft Drafteranthropics/claude-for-legal · Apache-2.0

Quick start

  1. 1Download and unzip. Download the skill ZIP from this page and unzip it into ~/.claude/skills/, or into a project's .claude/skills/ folder.
  2. 2Or upload in the Claude apps. In the Claude apps, open Settings, go to Capabilities, then Skills, and upload the ZIP.
  3. 3Confirm the plugin config exists. Make sure the regulatory-legal plugin is installed and its CLAUDE.md config is present at the plugin config path.
  4. 4Gather the gap and policy text. Have the GAP-ID or a gap description ready, plus the approved, latest policy text or a file path to it.
  5. 5Ask for the redraft. Say something like redraft the policy for a named gap, and answer any questions the agent asks about inputs.
  6. 6Review the marked-up draft. Read the proposed redraft file, check every [verify] tag, and confirm the rule is in force.
  7. 7Hand off for approval. Give the draft to the policy owner, who applies and approves the change. The gap closes only after that.

What is an agent skill, and what does policy-redraft do?

An agent skill is a folder containing a SKILL.md file with instructions that an AI agent loads when a request matches the skill's description. This one is named policy-redraft and is part of the claude-for-legal regulatory-legal plugin.

The skill starts after a gap has been identified, either with /regulatory-legal:gaps or /regulatory-legal:policy-diff, or described directly by you. The agent reads the current approved policy and the rule text, checks whether the rule is still in force, and proposes the smallest possible wording change, with inline notes explaining each change. The output goes to a new, clearly labeled draft file.

  • It never overwrites the original policy document.
  • It never marks the gap as closed in the tracker.
  • If it cannot verify that a rule is in force, it prints a RULE STATUS UNVERIFIED banner and tells you not to apply the redraft until the status is confirmed.
  • Changes use strikethrough for removed text and bold for inserted text, and each change carries a comment saying why it was made.

Install the policy-redraft skill

Download the ZIP for this skill from this page. The skill is distributed as a single file, SKILL.md, inside the package.

Option A, for Claude Code on your own machine: unzip the download into ~/.claude/skills/ so that the skill folder sits there. To make it available only inside one project, unzip it into that project's .claude/skills/ folder instead.

Option B, in the Claude apps: open Settings, go to Capabilities, then Skills, and upload the ZIP.

The skill expects the claude-for-legal regulatory-legal plugin to be installed, along with its config at ~/.claude/plugins/config/claude-for-legal/regulatory-legal/CLAUDE.md. Without that config, the skill cannot load your policy library index or practice profile.

Prepare the inputs the skill needs

The skill asks for three inputs and will ask you again if any is missing. It does not guess them.

The first is the gap. Either give a GAP-ID from your gap tracker file, gap-tracker.yaml, or describe the requirement, the regulation and the affected policy in your message, or paste a diff summary from /regulatory-legal:policy-diff.

The second is the current policy text. Give a file path or paste the text. If you give a file, the agent will ask whether it is the latest approved version, because a redraft against an outdated policy creates divergence. Pasted text is accepted but flagged in the reviewer note for you to confirm.

The third is the rule text. This can come from the diff output, a fetched regulation with its source noted, or text you paste, which is tagged as user provided. If the rule text is partial or ambiguous, the agent will offer options and wait for you to choose.

Trigger the skill with natural requests

You do not need to type the command. The skill is designed to load when you say things like the phrases in its description. You can also invoke it directly with the argument hint GAP-ID or gap description.

Good requests name the gap and the policy, and say what you want. The skill's description lists these trigger phrases: redraft the policy, draft the policy fix, and mark up the policy. A companion gap-finding step can also hand a gap over for drafting.

  • Redraft the policy for GAP-014 using the current data retention policy.
  • Draft the policy fix for the missing biometric identifiers gap from the last policy-diff.
  • Mark up the policy section that covers vendor notification for the new rule we found.

What you get back and what happens next

The agent writes a Policy Redraft Memo to a new file named after the policy with the word proposed-redraft and today's date, so the source document is never touched. If you use matters, output goes to the matter folder instead.

The gap stays open in the tracker. Applying and approving the change is the policy owner's job. When that is done, you tell the agent and it can update the tracker. If you ask it to apply the change to the policy or close the gap immediately, it will decline and explain why.

  • Review every [verify] tag before relying on a date, threshold or citation.
  • Check that the rule is still in force at the official docket or with outside counsel.
  • Keep the reviewer note with the draft when it goes to the policy owner.

Safety notes: scripts and network access

This skill does not run scripts. It does use the network: when a rule's status needs checking, the agent may use research tools or web search if you have them enabled, and it may check the Federal Register docket.

Treat anything fetched from the web as needing verification. The skill tags such material as web search, verify, and the redraft keeps those tags so they are not lost.

Tips for better redrafts

Give the most specific gap you can, with the exact requirement and the policy section. Smaller, clearer inputs produce smaller, clearer edits.

Always confirm you are working from the approved, latest policy version. The skill's smallest-possible-edit rule means it only changes sections tied to the gap, so unrelated sections should remain as they were. If other parts change, ask why.

What people use it for

  • Drafting a first-pass update to a data retention policy after a new regulatory rule is published.
  • Producing a reviewable redline of a vendor notification section that a policy-diff flagged as out of date.
  • Preparing a marked-up starting point for a policy owner who would otherwise start from a blank page.
  • Recording why each wording change was made, so reviewers can accept or reject it line by line.

Frequently asked questions

Is the policy-redraft skill free?

The source material for this page does not state pricing or licence terms beyond the Apache-2.0 licence listed for the repository. Check the licence and any Claude plan requirements yourself before relying on it.

What do I need to use policy-redraft?

You need Claude Code or a compatible agent with the claude-for-legal regulatory-legal plugin installed, a CLAUDE.md config at the plugin config path, and a gap (a tracker entry or a description) plus the approved policy text. Web search or research tools are optional, for verifying rule status.

Will policy-redraft edit my policy document directly?

No. The skill is designed to write to a new, clearly labeled proposed-redraft file and to refuse requests to apply changes to the source policy. The policy owner applies and approves any change.

Does the skill close the gap in my tracker?

No. The gap closes only after the redraft is applied and approved by the policy owner. The agent will update the tracker when you tell it that approval has happened.

Why does the skill say my rule status is unverified?

It checks for red flags such as a compliance date passed by more than 30 days, a rule more than 12 months old, or a contentious final rule. If it cannot confirm the rule is in force, it prints a banner telling you not to apply the redraft until the status is checked at the Federal Register docket or with outside counsel.

How does this compare with drafting the policy change myself?

The skill gives you a starting redline with a reason attached to each change, which saves drafting time. It does not replace legal judgement; the source says the output needs professional review. Source material does not compare it with other specific tools.

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-redraft if you are in-house legal or compliance staff, or a policy owner, who already works with the claude-for-legal regulatory-legal plugin and wants a reviewable first draft for a regulatory update. Pick something else if you want an agent to edit or apply policy changes directly, or if you have no qualified reviewer, because the output is explicitly a first draft that must be checked before use.

Download the skill