Skills

Skills are packaged instructions and rules that extend an AI agent's capabilities with Knock-specific procedural knowledge. They sit alongside your tool surfaces — the Knock CLI, the MCP server, and the Knock agent in the dashboard — and teach the agent how to use those tools to accomplish real work.

Once a skill is available to an agent, it auto-activates when you ask the agent to work on a related task. There's no manual prompting required.

Skills come from two places at Knock:

• The open-source knocklabs/skills package that you install into an AI coding agent (Cursor, Claude Code, and similar). These are the skills covered in Available skills below.
• Your own skills authored inside the Knock agent in the dashboard. The dashboard agent supports creating skills directly in your account and invokes them automatically during agent runs, which is useful for codifying team-specific conventions that aren't captured in the open-source package.

Knock publishes an open-source skills package on GitHub at github.com/knocklabs/skills. Install it into any compatible AI coding agent (Cursor, Claude Code, and similar) so the agent can use Knock's skills when working on your codebase.

Install the entire package with a single command:

Or install specific skills by passing a flag:

The Knock agent in the dashboard also supports skills. You can create skills directly in your Knock account and the dashboard agent invokes them automatically during agent runs.

Use dashboard skills to:

• Codify team-specific conventions that aren't captured in the open-source package (for example, your internal naming standards, copy guidelines, or workflow patterns).
• Standardize how the agent handles common requests across your account.
• Layer organizational context on top of Knock's built-in agent behavior.

The open-source package and dashboard-authored skills are complementary. A team might install notification-best-practices from the open-source package and also author an internal "transactional email tone of voice" skill in the dashboard.

The knock-cli skill teaches AI agents how to use the Knock CLI to manage your Knock resources from your codebase.

Complements. The Knock CLI.

Use it when you want an agent to:

• Pull workflows, templates, guides, and partials to your local machine.
• Push local changes back to Knock.
• Manage Knock resources as part of a development or CI/CD workflow.

The notification-best-practices skill gives AI agents comprehensive guidelines for notification design and implementation.

Complements. Any surface — the Knock CLI, the MCP server, and the Knock agent in the dashboard. Because this skill is about content and UX rather than how the agent reaches Knock, it applies regardless of which surface the agent uses.

Use it when you want an agent to:

• Write notification copy across channels (email, SMS, push, in-app).
• Apply best practices for transactional and welcome email templates.
• Follow channel-specific formatting and content guidelines.

Skills are not tied to a single tool surface. Different skills complement different surfaces, and many teams install or author more than one:

Skill	Pairs with	What it adds
knock-cli	The Knock CLI	How to scaffold, edit, and push resource files from your codebase.
notification-best-practices	Any surface (CLI, MCP, dashboard agent)	How to write effective notifications across channels.
Dashboard-authored skills	The Knock agent in the dashboard	Team-specific conventions and account-level context.

When an agent has both a tool surface and the right skill available, it can do more with fewer prompts: the tools give it reach into Knock, and the skill gives it the patterns to use those tools well.

• Knock AI overview. See how skills fit alongside the dashboard agent, agent function, CLI, and MCP server.
• Knock agent. The in-dashboard assistant where you can author and invoke skills directly.
• Knock CLI. The local-first agent surface that pairs with the knock-cli skill.
• MCP server. Connect external clients and your own apps to Knock as tools.