How to use the Is This a Problem agent skill

What is an agent skill and what does this one do?

An agent skill is a folder containing a SKILL.md file that tells an AI agent how to handle a specific kind of task. This skill, named is-this-a-problem, is written for quick legal questions of the kind that often arrive in Slack.

When you ask a question, the skill checks it against a risk calibration table your company has written. It then replies with one of three verdicts: Fine, Needs a look, or Hold. Each verdict comes with one sentence of reasoning and a next step.

The skill also watches for questions that sound simple but hide a trap. Examples in the source include customer logos on a website and adding a new data vendor. For these, it asks one clarifying question before answering.

• ✅ Fine: no problem, go ahead.
• ⚠️ Needs a look: routed for real review, with the work and timeline named.
• 🛑 Hold: stop and talk to someone before committing to a date or promise.

What do you need before installing the skill?

The source lists three requirements. You need Claude Code or a compatible agent runtime that loads SKILL.md files. You need a practice-level CLAUDE.md file at ~/.claude/plugins/config/claude-for-legal/product-legal/ that contains a Risk calibration section. Without that section the skill has no company-specific rules to apply.

Related skills in the same plugin, /ai-governance-legal:use-case-triage and /ai-governance-legal:vendor-ai-review, are used for routing when they are installed. They are optional for this skill to load, but the skill refers to them for certain trap cases.

How do you install the skill?

Download the ZIP from the skill page on this site. The skill's only file is SKILL.md. You can install it in one of two ways.

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

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

How do you trigger the skill with natural requests?

You can invoke the skill with its slash command, which takes the question as an argument. You can also just talk naturally. The skill's description lists phrases that should activate it, such as "is this a problem", "quick question", "can we do X", "do I need legal review for", and "sanity check". Pasting a colleague's question that needs a same-minute call also works.

The slash command format from the source is shown below.

• Say "is this a problem" followed by your question
• Say "quick question" or "sanity check" and paste the Slack message
• Ask "do I need legal review for" with the feature or change in question

What examples does the skill give for each verdict?

The source shows short example replies. A Fine reply names the pattern and says to ship it. A Needs a look reply names the work, such as a PIA, and asks whether to start it. A Hold reply stops the change and asks what the customer agreement says before anyone promises a date.

For questions that touch a trap, the skill asks one catch question first. For customer logos it asks what the contract says about publicity and whether written permission exists. For a new vendor it asks what data flows to them. For A/B tests it asks whether both arms see the same price for the same thing.

• Customer logos on the site: asks about publicity terms and written permission
• Adding a vendor to an integration: asks what data flows to them
• Training on existing data: asks what users were told at collection and where they are

What safety and privacy points should you know?

The skill runs no scripts and uses no network access, according to its source listing. It does, however, read your practice-level CLAUDE.md and, if matter workspaces are enabled, matter files in the matter folder.

The skill warns about destinations. If you name a public channel, a company-wide list, a counterparty, a vendor, or a client, it asks whether the destination is inside the privilege circle. Public channels and outside parties can waive legal protection, so the skill offers a privileged version for legal only, a sanitized version for a broader channel, or both. It does not silently add a privileged header and help you paste the output somewhere that header would not protect it.

• No scripts are run and no network calls are made
• Check the destination before sharing a reply outside legal
• Do not treat a Fine verdict as final legal advice on high-stakes matters