2 min lesson
The three invocation patterns and the runtime-agnostic interface
Compare two rows from "The three invocation patterns and the runtime-agnostic interface", then say when each one fits.
Step 1 of 2
The three invocation patterns and the runtime-agnostic interfacethe part candidates skip
The SDK is a single, runtime-agnostic interface: one set of code targets local, Cursor Cloud and self-hosted-cloud runtimes - you pass the runtime you want and the rest is identical. There are three core concepts (the agent object, the prompt interface and message/stream handling) and three invocation patterns worth naming explicitly because they map to different jobs.
- Pattern
- agent.prompt
- Shape
- Fire-and-forget one-shot
- Use it when
- A single discrete task: summarize a diff, run a check, draft a changelog entry.
- Pattern
- Durable agent
- Shape
- Back-and-forth conversation
- Use it when
- Multi-turn work where you keep talking to the same agent across steps.
- Pattern
- agent.resume
- Shape
- Pick up an existing agent
- Use it when
- Preserve context across surfaces - resume a run started elsewhere instead of rebuilding state.
| Pattern | Shape | Use it when |
|---|---|---|
| agent.prompt | Fire-and-forget one-shot | A single discrete task: summarize a diff, run a check, draft a changelog entry. |
| Durable agent | Back-and-forth conversation | Multi-turn work where you keep talking to the same agent across steps. |
| agent.resume | Pick up an existing agent | Preserve context across surfaces - resume a run started elsewhere instead of rebuilding state. |
Same harness behind all three; the choice is one-shot vs. conversation vs. resuming an existing run.
It has full model neutrality (specify any model per agent), supports subagents (passed into agent.create) for context and token reuse, and lets you inject MCPModel Context Protocol. A standard that lets an AI agent pull in context from outside the repo, like Jira tickets or internal docs. Press Enter for the full definition. servers, skills and hooks at runtime. TypeScript and Python are both documented today; the fastest start is the built-in /sdk skill, which scaffolds working code, plus the SDK cookbook repo. A REST API exists as an alternative.
The clean teaching line: a skill or slash command only works inside a Cursor surface; the SDK brings the same agent anywhere - any TypeScript service, CI pipeline or backend - so it is a superset of a skill. A vivid proof point: Cursor's field team built a security-questionnaire app on the SDK - upload a questionnaire doc and it spawns a cloud agent that does a first pass against the trust-center docs, privacy policies and security documentation. SDK use cases span finance, legal, support and design, not just coding.