Guide · Agent skill
How to use the Policy Monitor agent skill (policy-monitor)
Updated · written with AI from the project's own documentation and checked against our catalogue data
The policy-monitor skill makes an AI agent act as a compliance reviewer for an organisation's AI policy. It compares recent AI impact assessments, triage results and vendor reviews against the published policy and a use-case registry, and reports gaps. This guide explains what the skill does, how to install it, how to trigger it with plain requests, and what it needs before it can work.
Policy Monitoranthropics/claude-for-legal · Apache-2.0 Download the skill Quick start
- 1Download and unzip. Download the policy-monitor ZIP from the directory page and unzip it.
- 2Place the skill folder. Move the folder into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. You can also upload the ZIP under Settings → Capabilities → Skills in the Claude apps.
- 3Configure CLAUDE.md. Make sure ~/.claude/plugins/config/claude-for-legal/ai-governance-legal/CLAUDE.md contains your outputs folder path, policy commitments and use-case registry.
- 4Run a sweep. Ask for a policy sweep or run /ai-governance-legal:policy-monitor with no argument to scan outputs created since the last sweep.
- 5Check a planned practice. Describe the new AI practice in quotes after the command to see whether the policy covers, misses or conflicts with it.
- 6Acknowledge results. Review the REQUIRED and ADVISABLE gaps, then tell the agent the results are reviewed so the tracking stamps are updated.
What the policy-monitor skill makes the agent do
An agent skill is a folder of instructions that Claude follows when a matching request comes in. This one is built for legal and governance teams that maintain an internal AI policy and want to catch drift between that policy and how AI is actually used.
It works in two modes. In sweep mode, it reads recent outputs from a configured folder, such as AI impact assessments (AIAs), triage results and vendor reviews, and compares them with the policy. In direct-query mode, you describe a planned AI practice and it reports whether the policy covers, misses or conflicts with it.
- Sweep mode: classifies each gap as REQUIRED (the policy misrepresents current practice) or ADVISABLE (the policy is silent).
- Each gap comes with the current policy wording quoted and suggested replacement language drafted.
- Direct-query mode: reports covered, missing or conflicting, plus suggested language and a timing recommendation.
- It only reads files and drafts suggestions. It does not change the policy itself.
Install the policy-monitor skill
Download the ZIP for the skill from the directory page and unzip it. Then place the folder in one of two locations, depending on how you want to use it.
For a personal install, put it in ~/.claude/skills/. For a single project, put it in that project's .claude/skills/ folder. Alternatively, you can upload the ZIP in the Claude apps under Settings → Capabilities → Skills.
- Personal: ~/.claude/skills/
- Project only: .claude/skills/ inside the project folder
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
Set up the configuration file and outputs folder
The skill depends on a plugin configuration file at ~/.claude/plugins/config/claude-for-legal/ai-governance-legal/CLAUDE.md. It needs to hold the outputs folder path, your policy commitments and your use-case registry. Without the outputs folder, the skill can still run direct-query checks but cannot run the crawl sweep.
If the outputs folder is not yet configured, the skill points you to the cold-start-interview command in the same plugin set to set it up.
- Outputs folder path (where AIAs, triage results and vendor reviews are stored)
- AI policy document location (the skill reads the actual document, not only the summary)
- AI policy commitments section and use-case registry (approved, conditional and never use cases)
Trigger the skill with natural requests
You can invoke the skill with the slash command or by describing what you want. The skill is triggered by phrases such as 'policy sweep', 'does our AI policy cover this', 'we want to start doing X, does the policy need updating' and 'run the policy monitor'.
Run a sweep with no argument or with --sweep. To check a planned practice, pass a description of it as the argument.
- Sweep: run the command with no argument, or with --sweep
- Direct query: describe the proposed practice in quotes after the command
/ai-governance-legal:policy-monitor
/ai-governance-legal:policy-monitor "We want to use AI to automatically flag expense reports for review"Read the results and acknowledge them
In a sweep, each gap includes the quoted current policy text, a description of the gap and drafted language. Review each item with qualified counsel before adopting any wording, because the suggested language is drafted by the agent and is not legal sign-off.
The skill does not update its tracking stamps when it runs. Only after a human acknowledges the results does it update the last sweep date and the gap count in the configuration file. This means an unreviewed sweep cannot quietly suppress the same gaps next time.
- REQUIRED: the policy says something a recent output contradicts, or an approved use case affecting external parties has no coverage.
- ADVISABLE: the policy is silent or vague, but the practice is not in conflict.
- Use-case outputs not yet in the registry are flagged separately.
Safety notes and limits
According to the skill's description, it runs no scripts and uses no network access. It reads local files and writes only the tracking stamps after acknowledgment.
The skill cannot schedule itself. The plugin suggests setting a weekly reminder in your own scheduler to run the command. Automatic recurring runs need a separate scheduled-tasks integration that is not bundled with this plugin.
- No scripts are run and no network calls are made.
- Suggested language must be reviewed by qualified counsel.
- Weekly automatic sweeps require your own scheduler integration.
What people use it for
- Catching an AIA that approved an AI-assisted hiring screen while the policy still says AI is not used in employment decisions.
- Checking whether a vendor's data-use terms are reflected in the policy before a new AI tool goes live.
- Asking whether an AI feature for drafting customer support replies needs a more specific policy statement.
- Finding use cases that appear in approved outputs but are missing from the use-case registry.
Frequently asked questions
Is the policy-monitor skill free to use?
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 file in the repository for the terms that apply to you.
What do I need before I can run policy-monitor?
You need a written AI policy, the Claude for Legal plugin configuration file at the path given above, and access to an outputs folder containing AIAs, triage results and vendor reviews. Without the configuration file and outputs folder, the skill depends on those files and will not work fully.
Can policy-monitor change my AI policy automatically?
No. The skill only reads files and drafts suggested language. It updates its tracking stamps only after a human acknowledges the sweep results, and you still need to make any edits to the policy yourself.
Can the skill run every week on its own?
Not by itself. The skill cannot schedule runs. Weekly recurring runs need a separate scheduler integration that is not bundled with the plugin, so you would set up your own reminder or scheduled task to run the command.
Is the suggested policy wording ready to publish?
Not without review. The suggested language is drafted by the agent and must be checked by qualified counsel before it is adopted.
How is policy-monitor different from a general AI policy checklist?
This skill compares real outputs in your outputs folder, such as AIAs and vendor reviews, with your actual policy document and registry. It is designed to find drift between practice and policy rather than give a generic checklist. The source does not compare it with specific alternative 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
Policy-monitor suits legal or compliance teams that already keep a written AI policy, an outputs folder and the Claude for Legal configuration, and want a regular way to catch drift. It is not a fit for organisations without those files, for final legal sign-off, or for teams that need the tool to run on its own schedule without a separate integration.
Download the skill