Glossary
Skills and Plugins (AI) — Definition for Lawyers
A skill is a packaged, reusable instruction an AI tool loads for one task; a plugin bundles skills with tools, connectors and automations. How they differ, and governance.
On this page
A skill is a packaged, reusable instruction set that an AI tool loads only when a task calls for it, such as "triage this NDA against our playbook". A plugin is a larger bundle that combines several skills with the tools, connectors and automations a complete workflow needs. Both turn know-how that would otherwise be retyped into every prompt into something a firm can write once, review and share.
What a skill contains
In Anthropic's format a skill is a folder whose core is a single Markdown file, SKILL.md. It opens with a short YAML frontmatter block giving the skill a name and a description, followed by plain-language instructions. The folder can also hold reference material, such as a playbook or a template, and scripts the model may run.
The description does the routing. The tool keeps only the short descriptions in view and loads the full instructions when a request matches one, or when the user invokes the skill by name, often as a slash command. That is why a firm can install dozens of skills without filling the context window: the detail arrives only when needed.
A minimal example of the file:
---
name: nda-triage
description: Triage an incoming NDA against the firm's standard positions
and classify it as green, yellow or red. Use when asked to review,
screen or triage a non-disclosure agreement.
---
1. Read playbook.md in this folder before reviewing.
2. For each clause in the playbook, quote the NDA's corresponding text or
state "not found".
3. Classify each clause as matching, acceptable fallback or escalate.
4. Give an overall rating: green if all match, yellow if any fallback,
red if any escalation item.
5. Never mark a clause as matching without quoting it. Output is a draft
for lawyer review.
What a plugin adds
A plugin wraps skills with the machinery to act. The Claude for Legal plugins published in Anthropic's open-source claude-for-legal repository show the typical parts:
| Component | What it is | Example use |
|---|---|---|
| Skills | Instruction files for individual tasks | Contract review, diligence grid, demand letter |
| Commands | Skills exposed as slash commands | /plugin-name:review |
| Scheduled agents | Tasks that run on a timer | Renewal watcher, docket monitor |
| Hooks | Rules that fire before or after an action | Block a send until a human confirms |
| Connectors | Definitions pointing at MCP servers | Document management, e-signature, research databases |
| Shared context | A profile file other skills read | Jurisdictions, standard positions, escalation rules |
Other vendors use different names for similar ideas: custom GPTs, agents in office suites, extensions in legal platforms. The distinction that carries across is the same. A skill tells the model how to do a task; a plugin also gives it the means, which is where governance starts to matter.
How they relate to Projects and system prompts
A system prompt or Project instruction applies to every message in its scope. A skill applies only when triggered. In practice firms use both: a Project for the matter, with its documents and standing instructions, and skills for the repeatable tasks performed inside it. The Claude Projects playbook covers the first; the prompt library playbook shows how a firm's tested prompts become the raw material for the second.
Governance before installation
Because skills are text and plugins can carry code and connectors, review them as you would any software that reaches client data:
- Source. Who wrote it, and can you read every file it contains?
- Reach. Which connectors does it define, and with what permissions?
- Actions. What can it do without a human approving: send, file, delete, run scripts?
- Ownership. Who in the firm maintains the skill when the playbook or the model changes?
Third-party skills are also a route for prompt injection: a malicious instruction file is still an instruction file. The agent security guide covers the controls.
Where this leaves you
Take the one task you prompt for most often, write it as a skill with its playbook attached, and test it on documents where you know the right answer. Module 4 of the AI Lab for Lawyers course covers reusable instructions, prompt libraries and custom assistants.
Frequently asked questions
What is the difference between a skill and a system prompt?
A system prompt applies to every message in a conversation, Project or custom assistant. A skill applies only when a request calls for it, either because its description matches what you asked or because you invoked it directly. That makes skills better for many narrow, repeatable tasks that would clutter a single standing instruction, such as NDA triage, a disclosure schedule format or a citation check.
Are third-party legal plugins safe to install?
Not automatically. A plugin can include connectors to your document store or email and scripts that run on your machine, so it carries the same risks as any software that touches client data. Read what it contains, prefer sources you can inspect, check what permissions its connectors request, and keep any action that sends, files or deletes behind a human approval.
Put this into practice
AI Lab for Lawyers: a live, hands-on course for practising lawyers
4 live sessions (8 hours), recordings included, taught by Dr. Niklas Schmidt of Wolf Theiss. Rated 4.7/5 from 381 ratings; alumni from Norton Rose Fulbright, Paul Hastings, Pillsbury and Taylor Wessing.
Related reading
Tools
Claude for Legal Plugins Explained: The 12 Practice-Area Plugins and 20+ Connectors
Claude for Legal gives paid Claude users 12 open-source practice-area plugins, 150+ skills and 20+ connectors to iManage, NetDocuments and CoCounsel.
8 min read
Glossary
System Prompt — Definition for Lawyers
A system prompt is the standing instruction an AI model reads before your message. Where it lives in Claude, ChatGPT and Skills, a legal template, and its limits.
4 min read
Glossary
Model Context Protocol (MCP) — Definition for Lawyers
Model Context Protocol (MCP) is the open standard that lets Claude, ChatGPT and Copilot read your DMS or docket feed through one connector. What a server exposes and what to check.
4 min read
Playbooks
Build a Prompt Library for Your Practice: Structure, Governance and 12 Starter Prompts
Build a law firm prompt library lawyers will use: folder structure, a header for every prompt, a review cadence for model changes and 12 tested starter prompts.
11 min read
Playbooks
Set Up Claude Projects for a Matter: Instructions, Files and Memory
Set up a Claude Project for a legal matter in 30 minutes: sample instruction text, which files to upload, Skills, memory, Team sharing and three prompt flows.
10 min read
Guides
Agent Security for Lawyers: Prompt Injection, Data Leaks and Guardrails
AI agent security for law firms: prompt injection via hostile documents, the Grok Build and Cowork incidents, and the allowlists and gates that contain them.
11 min read