Skills, connectors, and plugins all extend what an AI agent can do, but they solve different problems. The simplest way to distinguish them is to ask three questions:
- A Skill answers: How should the agent perform this work?
- A connector answers: Which external information and actions can the agent access?
- A plugin answers: How can a complete capability be packaged, installed, and maintained?
They are not competing formats. A useful agent workflow often combines all three.
The short version
| Skill | Connector | Plugin | |
|---|---|---|---|
| Purpose | Package expertise and a repeatable method | Connect the agent to external systems | Distribute a complete set of capabilities |
| Think of it as | A playbook | A secure bridge | An installable toolkit |
| Typical contents | Instructions, references, templates, examples, scripts | Tools and data exposed through a defined interface | Skills, connector configurations, commands, hooks, scripts, and settings |
| Use it when | The agent needs to follow a consistent process | The agent needs current data or must act in another system | Several components should be installed and versioned together |
A Skill packages expertise
General-purpose agents are capable, but they do not automatically know your professional methodology, preferred structure, fallback positions, or quality standards. A Skill provides that missing guidance.
An Agent Skill is an organized folder of files that packages procedural knowledge. Its main file explains what the Skill is for and how the agent should approach the task. The folder can also include clause libraries, templates, examples, checklists, reference material, or small supporting scripts.
For example, an NDA review Skill might tell the agent to:
- Identify which party the user represents.
- Compare each clause with the firm's preferred positions.
- Flag unusual confidentiality exclusions and term lengths.
- Explain the practical risk of each deviation.
- Produce comments and proposed drafting in a consistent format.
The Skill does not need access to a remote system to be useful. It can work with a document the user provides. Its purpose is to make the agent's execution more consistent and closer to the method of a domain expert.
Because Skills are folders, they are easy to inspect and move. Teams can version them in Git, store them in a shared drive, adapt them to a particular practice, or zip and share them without adopting a new proprietary format.
A connector gives the agent access
A connector links an agent to information or actions outside the current conversation. On Lawve, connectors generally use the Model Context Protocol, or MCP, an open standard for connecting AI applications to external tools and data.
A connector might let an agent:
- Search legislation, case law, or regulatory decisions.
- Retrieve a document from SharePoint, Google Drive, iManage, or NetDocuments.
- Read matter information from a practice-management system.
- Create a task, update a record, or save an approved document.
The connector defines what the agent is allowed to request and how the external system responds. Authentication and permissions are managed at that boundary, so the agent only receives the access granted to it.
What a connector does not provide is the professional method. Access to a case-law database does not tell the agent which authorities are persuasive, how to assess conflicting decisions, or how to present the conclusion. That methodology belongs in a Skill.
The distinction is therefore simple: a Skill tells the agent what to do and how to do it; a connector gives it the information or actions needed to carry out the work.
A plugin packages a complete capability
A plugin is an installable and versioned package for distributing a broader capability. Depending on the AI application it targets, a plugin can bundle Skills, connector configurations, commands, hooks, scripts, and related settings behind one manifest.
Imagine a legal team wants to distribute a complete privacy-notice review workflow. Installing each part separately would require users to locate the right Skill, configure the correct legal-source connector, add supporting commands, and keep every component compatible as they evolve.
A plugin can assemble those pieces once and distribute them as one named version. Everyone who installs that version receives the same setup, and an update can be released as a new version with clear release notes.
A plugin is not a new form of intelligence. The underlying Skill still provides the method, and the connector still provides access. The plugin is the delivery and maintenance layer that keeps the pieces together.
Plugin formats can vary between AI applications, so compatibility matters. Before installing one, check which host it supports, what it contains, which external systems it connects to, and what permissions it requests.
How they work together
Consider an agent that reviews a supplier agreement and records the result in the firm's matter system.
The Skill contains the review methodology: the clauses to identify, the firm's preferred positions, the risk scale, and the expected output.
The connector retrieves the agreement and relevant matter information, searches an approved precedent library, and saves the final review to the correct record.
The plugin packages the Skill, connector configuration, commands, and supporting settings so the whole workflow can be installed consistently across the team.
Removing any one layer changes the result:
- Without the Skill, the agent has access but no reliable review method.
- Without the connector, the agent can follow the method but needs the user to provide and move the information manually.
- Without the plugin, the workflow can still work, but every user must assemble and maintain its components separately.