How to use the AI Use Case Triage agent skill (use-case-triage)

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

An agent skill is a folder containing a SKILL.md file that tells a compatible agent how to perform a specific task. This skill is built for in-house legal and compliance work: when someone asks whether a proposed AI use is allowed, the agent follows the skill's steps rather than improvising a general answer.

The skill reads your practice profile file, looks up the proposed use in your registry, checks it against your red lines, and classifies it. For conditional cases it lists the concrete conditions and next steps, and it flags where privacy or product counsel needs to be involved. If the use case is not in the registry, it defaults to CONDITIONAL pending a formal AI impact assessment rather than guessing.

• Output includes the classification, the reasoning, a conditions table (for conditional cases), a governance tier, and any cross-plugin handoffs.
• If the use case is missing from the registry, the agent proposes a registry update once it is settled.
• The triage is a first pass. The skill's own text says regulatory citations need verification and the output still needs lawyer review.

Before you install: what the skill requires

The skill needs Claude Code or a compatible agent that can read SKILL.md skills. It also needs a configured practice profile at ~/.claude/plugins/config/claude-for-legal/ai-governance-legal/CLAUDE.md that contains your use case registry and red lines, and the claude-for-legal ai-governance-legal plugin, because the cold-start-interview and aia-generation skills are referenced.

If the profile has not been configured, the skill stops and points you to setup. You can instead say "provisional" to run against generic defaults (US jurisdiction, middle risk appetite, lawyer role, no playbook). Provisional output is tagged [PROVISIONAL] throughout.

Install the use-case-triage skill

Download the ZIP from the skill's page on this site and unzip it. The skill is a single file, SKILL.md. Place the unzipped folder in one of these locations:

For a personal install available across your projects, put it in ~/.claude/skills/. For a single project, put it in that project's .claude/skills/ folder. Alternatively, in the Claude apps you can upload the skill under Settings → Capabilities → Skills.

• Personal: ~/.claude/skills/
• Project only: the .claude/skills/ folder inside your project
• Claude apps: Settings → Capabilities → Skills (upload)

Trigger the skill with natural requests

You do not need to memorise a command. The skill's description lists phrases that activate it. Say things like "triage this use case", "can we use AI for X", "is this approved", or "what do we need to do to use AI for X".

The skill also accepts an argument hint: describe the use case, or write 'batch' to triage a list. The slash form documented in the skill is shown below.

• Give the vendor or tool, what the AI does, who it acts on, and whether a human reviews output before anything happens.
• Say whether the use is internal only or touches customers or third parties.
• If your description is vague, the agent asks follow-up questions before classifying.

Read the output: approved, conditional, or not approved

APPROVED means the registry allows the use. CONDITIONAL means the use may proceed once listed conditions are met, and the agent gives concrete conditions and next steps. NOT APPROVED means the use falls under a red line or is otherwise ruled out by the registry.

If the use case is a near match to a registered entry, the agent says so and notes that a materially different scope may need its own assessment. Any regulatory citation in the output is tagged by source and verification level, such as [settled], [verify], or [verify-pinpoint], and should be checked against a primary source.

• Direct match: the registry entry is applied.
• Near match: the agent flags the difference in scope.
• No match: CONDITIONAL pending an AI impact assessment, with a preliminary risk read.

Safety notes: scripts and network access

This skill runs no scripts and uses no network access, according to its listing. Its only file is SKILL.md. The skill reads the local practice profile and, if you have the plugin, can route to other skills, but it does not itself make outside calls.

Matter-specific output is written to a matter folder under ~/.claude/plugins/config/claude-for-legal/ai-governance-legal/matters/, only if matter workspaces are enabled in your profile. The skill does not read other matters' files unless cross-matter context is switched on.

Tips for getting reliable results

The quality of the triage depends on the registry. Populate it and your red lines before relying on the output. Keep the jurisdictions in your profile accurate, because the skill applies only the regulatory footprint configured there and flags other jurisdictions for re-triage rather than extending by analogy.

Give the agent specifics. A request such as "sales wants AI to score leads" triggers clarifying questions; adding the vendor, data sources, and whether a human reviews the scores gets you a classification faster.

• Configure the profile first, or accept that provisional output reflects generic defaults.
• Treat conditions and dates as items to confirm with counsel.
• Use the output as a routing step toward a full impact assessment for anything conditional.